[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7FDWTKH0VR0BA77KSWQG1Q9":3,"skill-same:01M7FDWTKH0VR0BA77KSWQG1Q9":55},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":24,"commitSha":25,"skillDirectory":26,"forksCount":27,"repositoryHtmlUrl":28,"repositoryHomepageUrl":6,"ownerHtmlUrl":29,"ownerBlogUrl":6,"ownerDescription":30,"updatedAt":31,"install":32,"resources":45},"01M7FDWTKH0VR0BA77KSWQG1Q9","autoreview",null,"Structured Codex, Claude, Amp, Pi, or Kimi code review when explicitly requested.","lza6\u002FArmor-piercing-system-prompts","lza6","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F128880206?v=4","01M7EB3K7WPX5T28EJ2NVHE6EG","Chinese-tingfeng","01M7EB3K97E1KQ16QWKJMFNGX8",11,"MIT","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fautoreview\u002FSKILL.md",[],[],"2026-11-16 10:05:34",false,"---\nname: autoreview\ndescription: \"Structured Codex, Claude, Amp, Pi, or Kimi code review when explicitly requested.\"\n---\n\n# Auto Review\n\nRun the bundled structured review helper only when the user explicitly asks for autoreview, a second-model review, or one of its named review engines. This is code review, not Guardian `auto_review` approval routing.\n\nCodex review is the default when no engine is set. It uses `gpt-5.6-sol` with `high` reasoning by default, then retries once with `gpt-5.6-terra` only when the account cannot access Sol. Claude review is optional and uses `claude-fable-5` by default. Amp review is optional and uses `openai\u002Fgpt-5.6-sol` with `high` reasoning by default. Pi and Kimi use the model configured by their respective CLIs unless `--model` overrides it.\n\nDo not invoke Autoreview automatically before a commit, push, PR, merge, deploy, or final reply. Repository or workflow rules may call it only when they explicitly name it.\n\n## Contract\n\n- Default accepted findings are P0 only: report issues worth blocking the current change\n  because they materially break the normal flow, outcome, or safety boundary.\n  Use `--max-priority P1`, `P2`, or `P3` only when the caller explicitly asks\n  for a wider review.\n- Treat review output as advisory. Never blindly apply it.\n- Verify every finding by reading the real code path and adjacent files.\n- Read dependency docs\u002Fsource\u002Ftypes when the finding depends on external behavior.\n- Reject unrealistic edge cases, speculative risks, unrelated rewrites, and fixes that over-complicate the codebase.\n- Prefer root-cause fixes at the right ownership boundary. A coherent refactor is appropriate when it removes the bug class, duplicate policy, stale paths, or ownership confusion; do not default to a symptom patch.\n- When an accepted finding exposes a bug class or repeated pattern, inspect its owner and relevant sibling implementations before fixing.\n- Fix the same bug class across its owner-boundary neighborhood when practical; stop at unrelated invariants, different owners, and unapproved contract changes.\n- Run one bounded review pass. If an accepted finding changes code, run the smallest relevant test; rerun Autoreview only when the user explicitly requests another pass.\n- For security-audit suppression changes, verify accepted findings remain auditable: suppressed findings stay in structured output, active output keeps an unsuppressible suppression notice, and aggregate findings cannot hide unrelated active risk.\n- Never switch or override the requested review engine\u002Fmodel except for the documented Codex Sol-to-Terra account-access fallback. Capacity, rate-limit, and unrelated failures keep the same engine\u002Fmodel.\n- Be patient with large bundles. Structured review can take up to 30 minutes while the model call is active, especially with Codex tools or web search.\n- Treat heartbeat lines like `review still running: ... elapsed=... pid=...` as healthy progress, not a hang. Let the helper continue while heartbeats are advancing. Pass `--stream-engine-output` when live engine text is useful; Codex and Claude filter tool\u002Ffile chatter, other runnable engines pass raw output through.\n- Do not kill a review just because it has been quiet for 2-5 minutes, or because it is still running under the 30-minute window. Inspect the process only after missing multiple expected heartbeats, after 30 minutes, or after an obviously failed subprocess; prefer letting the same helper command finish.\n- Tools are useful in review mode. Codex receives the validated bundle in an empty workspace so ignored files and linked-worktree metadata remain unreadable; web search stays available for dependency contracts and upstream docs.\n- Security perspective is always included, but it should not cripple legitimate functionality. Report security findings only when the change creates a concrete, actionable risk or removes an important safety check.\n- Reviewer subprocesses preserve engine authentication and non-credentialed proxy variables needed by headless or restricted-network environments while stripping process-injection, Git override, and credentialed proxy values.\n- Credential checking belongs to the reviewer: the hard rules require inspection of the entire change bundle for accidentally committed credentials and require every suspected real credential to be reported as a P0 security finding without reproducing its value. The helper does not invoke an external credential scanner. Security-sensitive paths remain omitted. Safe large diffs are sent as one pass while they fit the aggregate prompt limit, then partitioned into complete bounded passes without truncation.\n- Regression provenance needs patch proof, not blame alone. `git log -S\u002F-G`, `git blame`, commit subjects, and PR metadata locate candidates. Before saying `introduced by`, inspect raw parents with `git --no-replace-objects cat-file -p \u003Csha>` and verify the implicated behavior changed in `git --no-replace-objects diff --no-ext-diff --no-textconv \u003Craw-parent> \u003Csha> -- \u003Cpath>`; a genuine root needs raw-header proof that it has no parents.\n- Blame `^sha`, porcelain `boundary`, and shallow\u002Fgrafted history alone are not introduction proof. `--root` can hide boundary markers; `git show` or `rev-list --parents` can make a shallow boundary look like a root. An available raw parent permits explicit comparison even at a shallow boundary; missing parents or an unverifiable patch require `unknown` with the gap. Use `carried forward` only for verified preexisting behavior and `made visible` only for a verified trigger. Apply the same bar to finding prose, summaries, and owner hints.\n- Keep code author, introducing PR author, merger, committer, automation trigger, and current PR author separate; none of those roles alone proves causation. Cite the verified commit\u002FPR\u002Fdate. If no PR is traceable, use the verified commit and known author identity; unknown identities stay unknown, and missing PR metadata is not a separate finding.\n- For automation merges, identify the human trigger only from explicit timeline\u002Fcomment\u002Fevent evidence, such as a maintainer automerge command or arming label. Report `automerge triggered by @login` only when verified; otherwise say trigger unknown. Triggering or merging is not proof of authorship or introduction.\n- Do not invoke built-in `codex review`, nested reviewers, or review panels from inside the review. The helper builds one validated bundle, calls the selected engine once for normal inputs or once per complete bounded chunk for oversized inputs, validates the structured results, and stops.\n- Stop as soon as the helper exits 0 with no accepted\u002Factionable findings. Do not run an extra review just to get a nicer \"clean\" line, a second opinion, or clearer closeout wording.\n- Treat `scoped-clean` with exit 0 as clean only for the selected Git target and requested priority. `filtered` is not a correctness certificate; `incomplete` requires resolving the scope mismatch or missing required finding before claiming clean.\n- If rejecting a finding as intentional\u002Fnot worth fixing, add a brief inline code comment only when it explains a real invariant or ownership decision that future reviewers should know.\n- If `gh`\u002FGitcrawl reports `database disk image is malformed`, run `gitcrawl doctor --json` once to let the portable cache repair before retrying review; do not bypass the shim unless repair fails and freshness requires live GitHub.\n- If Gitcrawl reports a portable manifest mismatch, source\u002Fruntime DB health error, or stale portable-store checkout, run `gitcrawl doctor --json` and inspect `source_db_health`, `runtime_db_health`, and `portable_store_status` before falling back to live GitHub.\n- Do not push just to review. Push only when the user requested push\u002Fship\u002FPR update.\n\n## Scope\n\nAutoreview does not expand the task. Fix only verified blockers in the requested path. Mention unrelated findings without opening a new workstream, and stop when the requested review pass is complete.\n\n## Skill Path (set once)\n\nSet the skill script paths once, then use `\"$AUTOREVIEW\"` and `\"$AUTOREVIEW_HARNESS\"` in the examples below.\n\nChoose one:\n\n```bash\n# Project-local skill in the current repo for Codex and other agents:\nexport AUTOREVIEW=\".agents\u002Fskills\u002Fautoreview\u002Fscripts\u002Fautoreview\"\nexport AUTOREVIEW_HARNESS=\".agents\u002Fskills\u002Fautoreview\u002Fscripts\u002Ftest-review-harness\"\n```\n\n```bash\n# Claude Code project-local skill in the current repo:\nexport AUTOREVIEW=\".claude\u002Fskills\u002Fautoreview\u002Fscripts\u002Fautoreview\"\nexport AUTOREVIEW_HARNESS=\".claude\u002Fskills\u002Fautoreview\u002Fscripts\u002Ftest-review-harness\"\n```\n\n```bash\n# Source checkout of openclaw\u002Fagent-skills:\nexport AUTOREVIEW=\"skills\u002Fautoreview\u002Fscripts\u002Fautoreview\"\nexport AUTOREVIEW_HARNESS=\"skills\u002Fautoreview\u002Fscripts\u002Ftest-review-harness\"\n```\n\n```bash\n# Global skill:\nexport AGENTS_HOME=\"${AGENTS_HOME:-$HOME\u002F.agents}\"\nexport AUTOREVIEW=\"$AGENTS_HOME\u002Fskills\u002Fautoreview\u002Fscripts\u002Fautoreview\"\nexport AUTOREVIEW_HARNESS=\"$AGENTS_HOME\u002Fskills\u002Fautoreview\u002Fscripts\u002Ftest-review-harness\"\n```\n\nWhen using Claude Code, set `AGENTS_HOME=\"$HOME\u002F.claude\"` for global skills.\n\nOn native Windows, choose the matching pair:\n\n```powershell\n# Project-local skill in the current repo for Codex and other agents:\n$AUTOREVIEW = \".agents\\skills\\autoreview\\scripts\\autoreview\"\n$AUTOREVIEW_HARNESS = \".agents\\skills\\autoreview\\scripts\\test-review-harness.ps1\"\n```\n\n```powershell\n# Claude Code project-local skill in the current repo:\n$AUTOREVIEW = \".claude\\skills\\autoreview\\scripts\\autoreview\"\n$AUTOREVIEW_HARNESS = \".claude\\skills\\autoreview\\scripts\\test-review-harness.ps1\"\n```\n\n```powershell\n# Source checkout of openclaw\u002Fagent-skills:\n$AUTOREVIEW = \"skills\\autoreview\\scripts\\autoreview\"\n$AUTOREVIEW_HARNESS = \"skills\\autoreview\\scripts\\test-review-harness.ps1\"\n```\n\n```powershell\n# Global skill:\n$AgentsHome = if ($env:AGENTS_HOME) { $env:AGENTS_HOME } else { Join-Path $HOME \".agents\" }\n$AUTOREVIEW = Join-Path $AgentsHome \"skills\\autoreview\\scripts\\autoreview\"\n$AUTOREVIEW_HARNESS = Join-Path $AgentsHome \"skills\\autoreview\\scripts\\test-review-harness.ps1\"\n```\n\n## Pick Target\n\nDirty local work relative to HEAD:\n\n```bash\n\"$AUTOREVIEW\" --mode local\n```\n\nWithout `--base`, this reviews HEAD-to-index, index-to-working-tree, and validated\nuntracked files only. `--mode auto` selects the same HEAD-based scope when dirty;\nit does not include the committed PR merely because the checkout is on a PR\nbranch. `--mode uncommitted` is an alias for `--mode local`. A clean local checkout\nwithout an explicit base has no local patch to review.\n\nTo review a dirty candidate against an explicit base, including a resolved merge\nthat has not been committed:\n\n```bash\n\"$AUTOREVIEW\" --mode local --base origin\u002Fmain\n```\n\nThe helper pins that base to a commit at target selection. It reviews base-to-index\nchanges and index-to-working-tree changes separately, plus validated untracked\nfiles. Only files identical across the base, index, and working tree are outside\nthe change bundle; staged changes later undone remain included. Actual binary\nor submodule changes still refuse review. The bundle labels its pinned\nstaged base. Git status remains relative to HEAD and does not define review scope.\nWithout `--base`, local mode retains its usual HEAD-to-index behavior; it never\ninfers a base from an in-progress merge.\n\nCommitted-only branch\u002FPR work:\n\n```bash\n\"$AUTOREVIEW\" --mode branch --base origin\u002Fmain\n```\n\nBranch mode reviews `BASE...HEAD` (merge-base-to-HEAD); staged, unstaged, and\nuntracked changes are excluded even in a dirty checkout. To review the **complete\nPR candidate including dirty rewrites**, pin the PR merge base and use local mode:\n\n```bash\npr_base=$(gh pr view --json baseRefName --jq .baseRefName)\nmerge_base=$(git merge-base HEAD \"origin\u002F$pr_base\")\n\"$AUTOREVIEW\" --mode local --base \"$merge_base\"\n```\n\nThe remote base must already be available and current locally; the helper does\nnot fetch. The pinned merge base avoids including unrelated upstream changes.\nAdd `--max-priority P2` when the caller requests P2 findings. An explicit `--base`\nalso applies when auto selects local, but explicit local mode avoids changing\ntargets when the checkout becomes clean.\n\nOptional review context is first-class. Prompt files and datasets must be repo-relative so review bundles cannot pull arbitrary host files. Context never expands the selected Git target, even if it contains a complete candidate diff or asks to review the whole PR. Finding membership is checked by changed file, not individual hunk:\n\n```bash\n\"$AUTOREVIEW\" --mode branch --base origin\u002Fmain --prompt-file review-notes.md --dataset evidence.json\n```\n\nIf an open PR exists, use its actual base:\n\n```bash\nbase=$(gh pr view --json baseRefName --jq .baseRefName)\n\"$AUTOREVIEW\" --mode branch --base \"origin\u002F$base\"\n```\n\nCommitted single change:\n\n```bash\n\"$AUTOREVIEW\" --mode commit --commit HEAD\n```\n\nUse commit review for already-landed or already-pushed work on `main`. Reviewing\nclean `main` against `origin\u002Fmain` is usually an empty diff after push. For a\nsmall stack, review each commit explicitly or review the branch before merging\nwith `--base`. Commit review compares the raw recorded parent with the selected\ncommit, ignoring replacement refs and legacy grafts. A genuine root compares\nagainst the empty tree. Missing parent objects stop review: explicitly deepen\u002Ffetch the needed\nhistory and rerun. The helper does not fetch it automatically.\n\n## Oversized Bundles\n\nThe helper validates the complete patch before partitioning it. A bundle that fits\nthe aggregate prompt limit remains one integrated review pass. Larger bundles\nare split at bundle sections and file boundaries where possible; an oversized\nsingle-file block is split at line boundaries with repeated file\u002Fhunk context\nand an absolute new- or old-file line offset. Untracked snapshots use\ninjection-safe source-line records so continuation passes retain reportable\nlocations. A single physical diff line split across passes also retains its\noriginal addition, deletion, or context marker.\nPrompt instructions remain whole in every pass. Large datasets are grouped from\ntheir validated file records, never by reparsing headings inside evidence.\nIndividual oversized datasets split at lines or UTF-8 boundaries with their\noriginal path and byte offset. Each evidence batch is paired with the complete\nchange bundle: every original change byte appears exactly once per evidence\nbatch, and every evidence byte is retained. All validated reports are merged\nbefore required-finding and exit-status checks.\nThere is no fixed pass-count ceiling: the complete frozen input determines the\nfinite pass sequence. The helper prints its size and pass count before running\npasses serially. Each pass retains the same prompt-size limit and\nreviewer isolation; a failed pass aborts without publishing a partial verdict.\n\nPreparation prints immediate phase updates and periodic elapsed time to stderr,\nwith file\u002Fbyte counts during hashing and no filenames or contents. These updates\nare separate from provider heartbeats and do not count against engine deadlines.\nBundle construction captures finding membership alongside validated text; normal\nand dry runs reuse that record without reopening untracked files for membership.\n\nDry runs reuse captured inputs and prompt construction without whole-tree integrity sweeps. Real\nreviews retain full fresh tree hashing before bundle construction, before review,\nand before publication, including unrelated tracked files, nonignored untracked files, index\nstate, and initialized submodules. Explicit prompt files and datasets also retain\ntheir own frozen bytes and raw path identities, regardless of Git ignore status\nor finding scope. They are revalidated before sending each pass and before\npublication; content changes, replacements, and leaf or ancestor symlink swaps\nrefuse stale results. These endpoint checks are not atomic filesystem snapshots\nand cannot guarantee detection of transient changes restored between checks.\n\nEvidence batches can multiply the pass count. Chunking cannot give one model\ncall every cross-file implementation detail. For architecture-heavy changes, still prefer\na coherent branch or PR shape whose semantic decision surface fits one pass.\nRemoving verified non-authoritative generated noise remains useful, but never\ndrop lockfiles, generated clients, policies, manifests, schemas, or other\nindependently semantic artifacts merely to shrink the review.\n\n## Models and thinking\n\nThe helper accepts `--model` globally or per engine (`engine=model`) and `--thinking` globally or per engine (`engine=level`). Repeat either flag for multiple reviewers.\n\nRecommended model defaults:\n\n| Engine              | Default model                                      | Source note                                           |\n| ------------------- | -------------------------------------------------- | ----------------------------------------------------- |\n| **codex** (default) | `gpt-5.6-sol` -> `gpt-5.6-terra` on access failure | OpenClaw org review default                           |\n| **claude**          | `claude-fable-5`                                   | Anthropic's most capable widely released Claude model |\n| **amp**             | `openai\u002Fgpt-5.6-sol`                               | Amp structured-generation review default              |\n\nCLI flags and environment variables override these defaults. Amp model IDs must use `provider\u002Fmodel` form. Pi and Kimi do not get built-in model defaults because their configured model catalogs may vary by installation.\n\n| Engine              | Model flag                 | Example model IDs                                                            | Thinking flag                             | Accepted levels                                            |\n| ------------------- | -------------------------- | ---------------------------------------------------------------------------- | ----------------------------------------- | ---------------------------------------------------------- |\n| **codex** (default) | `codex --model X exec ...` | `gpt-5.6-sol`, then `gpt-5.6-terra` on Sol access failure                    | `-c model_reasoning_effort=Y`             | `none`, `minimal`, `low`, `medium`, `high`, `xhigh`, `max` |\n| **claude**          | `claude --model X`         | `claude-fable-5`, `claude-opus-4-8`, `claude-sonnet-4-6`, `claude-haiku-4-5` | `--effort Y`                              | `low`, `medium`, `high`, `xhigh`, `max`                    |\n| **amp**             | Amp `amp.ai.generate`      | `openai\u002Fgpt-5.6-sol`                                                         | `reasoningEffort`                         | `none`, `low`, `medium`, `high`, `xhigh`, `max`            |\n| **pi**              | `pi --model X`             | `anthropic\u002Fclaude-sonnet-4`, `openai\u002Fgpt-4o`                                 | `--thinking Y`                            | `off`, `minimal`, `low`, `medium`, `high`, `xhigh`         |\n| **kimi**            | `kimi --model X`           | A model alias from the user's Kimi config                                    | `[thinking] enabled` in the staged config | `on`, `off`                                                |\n\nClaude also supports `--fallback-model a,b` for availability-based fallback chains ([model-config](https:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fmodel-config)). Current Claude docs note that auth, billing, rate-limit, request-size, and transport errors do not trigger fallback, and the changelog documents interactive-session support in `v2.1.166`.\n\n[OpenAI's model guidance](https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Flatest-model) identifies Sol as the GPT-5.6 frontier-capability route and documents `max` support. Autoreview keeps `high` as its default; use `max` only for the hardest quality-first reviews after comparing its latency and cost with `xhigh` on representative changes.\n\nExamples matching current `main` behavior:\n\n```bash\n# Codex with explicit model and reasoning\n\"$AUTOREVIEW\" --engine codex --model gpt-5.6-sol --thinking high\n\n# Codex fast mode (priority service tier); needs a model whose catalog lists the tier, silently standard otherwise\n\"$AUTOREVIEW\" --engine codex --codex-speed fast\n\n# Safe Codex model\u002Fresponse tuning overrides (--codex-speed wins over a service_tier here)\n\"$AUTOREVIEW\" --engine codex --codex-config 'service_tier=\"fast\"'\n\n# Claude Code aliases or full model names, with optional availability fallback\n\"$AUTOREVIEW\" --engine claude --model claude-fable-5 --thinking max\n\"$AUTOREVIEW\" --engine claude --model claude-fable-5 --fallback-model claude-opus-4-8,claude-sonnet-4-6\n\n# Amp direct structured generation (requires AMP_API_KEY)\n\"$AUTOREVIEW\" --engine amp --model openai\u002Fgpt-5.6-sol --thinking high --amp-bin amp\n\n# Pi with explicit model and thinking level\n\"$AUTOREVIEW\" --engine pi --model anthropic\u002Fclaude-sonnet-4 --thinking high --pi-bin pi\n\n# Kimi with its configured default model, or a configured model alias\n\"$AUTOREVIEW\" --engine kimi --thinking on --kimi-bin kimi\n\"$AUTOREVIEW\" --engine kimi --model kimi-model-alias\n\n```\n\n### Environment defaults\n\nCLI flags take precedence over environment variables.\n\nStore persistent personal defaults in your shell startup file or launcher\nenvironment. For repository-local defaults, use an existing local environment\nloader such as an untracked `.envrc`; the helper does not write a config file.\n\n| Variable                            | Purpose                                                                                                                          |\n| ----------------------------------- | -------------------------------------------------------------------------------------------------------------------------------- |\n| `AUTOREVIEW_MODEL`                  | Override the built-in default `--model` for all engines                                                                          |\n| `AUTOREVIEW_THINKING`               | Default `--thinking` for all engines                                                                                             |\n| `AUTOREVIEW_FALLBACK_MODEL`         | Default Claude `--fallback-model` chain                                                                                          |\n| `AUTOREVIEW_ENGINE_TIMEOUT_SECONDS` | Optional positive wall-clock limit for each reviewer process; disabled by default                                                |\n| `AUTOREVIEW_\u003CENGINE>_MODEL`         | Per-engine model override, for example `AUTOREVIEW_CODEX_MODEL=gpt-5.6-sol`                                                      |\n| `AUTOREVIEW_\u003CENGINE>_THINKING`      | Per-engine thinking override                                                                                                     |\n| `AUTOREVIEW_CODEX_CONFIG`           | Safe Codex model\u002Fresponse tuning overrides, semicolon-separated, e.g. `service_tier=\"fast\"`; capability-bearing keys fail closed |\n| `AUTOREVIEW_CODEX_SPEED`            | Codex service tier override: `fast` (priority), `flex`, or `default`; silently standard when the model does not list the tier    |\n| `AUTOREVIEW_CLAUDE_FALLBACK_MODEL`  | Claude-only fallback chain                                                                                                       |\n| `AUTOREVIEW_PROVIDER_ENV_ALLOW`     | Comma-separated custom Pi credential variable names; names must end in a recognized credential suffix                            |\n| `AMP_API_KEY`                       | Required Amp API credential; file\u002Fkeychain auth is intentionally excluded from the isolated runtime                              |\n\nCodex maps thinking to `model_reasoning_effort`. Claude maps thinking to `--effort`. Amp maps thinking to `amp.ai.generate.reasoningEffort`. Pi maps thinking to `--thinking`. Kimi maps `on` and `off` to `[thinking] enabled` in the staged review config. Only Claude accepts `--fallback-model`; global CLI\u002Fenv fallback requires at least one Claude reviewer, and engine-specific fallback overrides require that reviewer to be selected. Non-Claude fallback overrides, including `AUTOREVIEW_\u003CNONCLAUDE>_FALLBACK_MODEL`, fail closed instead of being silently ignored.\n\nAmp receives only `AMP_API_KEY` from the caller. Autoreview intentionally ignores `AMP_URL`, user settings, stored authentication, inherited MCP configuration, and other runtime variables. The API key's authenticated account and workspace must have no personal or workspace plugins: current normal Amp execution loads every authenticated plugin, so the preflight requests the complete inventory and fails before creating the review prompt unless the generated adapter is the only plugin. A dedicated Amp API key\u002Faccount without plugins is the safest setup. Amp can still discover personal and workspace skill metadata, but the isolated settings deny every local and remote MCP server before use, and the outer adapter has no skill tool. Custom Amp endpoints are not supported because forwarding an arbitrary endpoint could disclose the API key and review bundle. Native Windows is refused because its `chmod` behavior cannot establish or attest the POSIX private-file permissions used here; use Linux, macOS, or WSL.\n\n## Review engine isolation\n\nWhen autoreview runs inside the repository under review, external reviewer CLIs must not load project-local trust or configuration that the branch controls.\n\n| Engine     | Isolation flags                                                                                                                                                                                                                                                                                               | Reference                                                                      |\n| ---------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------ |\n| **codex**  | Auth-only config overrides, isolated workspace, `exec --ignore-user-config --ignore-rules --skip-git-repo-check`, plus read-only sandbox                                                                                                                                                                      | Codex CLI `exec --help`                                                        |\n| **claude** | `--safe-mode --setting-sources user --strict-mcp-config --disallowedTools mcp__*`; auto-memory and filesystem\u002Fshell tools disabled; empty external workspace; WebSearch by default (`v2.1.169+`)                                                                                                              | Claude Code [CLI reference](https:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fcli-reference)     |\n| **amp**    | Empty external workspace and isolated HOME\u002FXDG roots; complete authenticated plugin inventory must contain only the generated adapter; catch-all MCP denial with a process-spawn probe; fixed outer trigger and one input-free adapter tool; private prompt only reaches schema-constrained `amp.ai.generate` | Amp [plugin API](https:\u002F\u002Fampcode.com\u002Fmanual\u002Fplugin-api) and local CLI `--help` |\n| **pi**     | `--no-approve --no-session --no-context-files --no-extensions --no-skills --no-prompt-templates --no-themes --no-tools`                                                                                                                                                                                       | Pi CLI `--help`; requires Pi `v0.79.0+`                                        |\n| **kimi**   | Empty external workspace; staged `KIMI_CODE_HOME` with sanitized config; Markdown custom agent with no tools\u002Fsubagents; explicit empty `--skills-dir`; isolated runtime state                                                                                                                                 | Kimi Code CLI `--help`; requires Kimi `v0.30.0+`                               |\n\nCodex `--ignore-user-config` skips config loading for the exec run. Autoreview reconstructs only the documented `cli_auth_credentials_store`, `forced_login_method`, and `forced_chatgpt_workspace_id` settings from `CODEX_HOME\u002Fconfig.toml`, keeping authentication usable without forwarding unrelated user configuration. Codex runs in an empty temporary workspace: the validated bundle is its sole repository input, ignored files and linked-worktree metadata remain unreadable, and the zero project-doc budget keeps workspace instructions out of the prompt. `--ignore-rules` skips user\u002Fproject execpolicy rules. Claude `--safe-mode` disables project hooks, skills, plugins, MCP servers, and CLAUDE.md; autoreview supplies WebSearch by default, permits only explicitly domain-constrained WebFetch rules, and exposes no filesystem or shell tools. Amp runs its local CLI with isolated HOME\u002FXDG roots and one generated adapter plugin whose name includes a fresh 128-bit random suffix. Current normal Amp execution loads all authenticated plugins, so the preflight deliberately requests that same complete inventory and fails before writing the private prompt unless the generated adapter is the only active plugin, with exactly its expected tool, agent, and mode. Users with personal or workspace plugins must use a dedicated plugin-free Amp API key\u002Faccount. Isolated `amp.mcpPermissions` reject every local command and remote URL. Before writing the private prompt, autoreview creates a temporary skill whose MCP command would write a marker, runs `amp tools list`, and requires Amp to report the policy rejection without creating the marker; it removes that skill before continuing. A custom outer mode then receives only a fixed harmless trigger and exposes exactly one trusted, input-free `autoreview_generate` tool. That tool reads the private prompt file and calls `amp.ai.generate` directly with an explicit system prompt and report schema, so the untrusted patch never enters the outer agent context. Autoreview requires the stream's leading init event to attest the empty working directory, the exact singleton adapter-tool inventory, and `mcp_servers: []`; it then requires exactly one correctly ordered empty-input tool call\u002Fresult and one terminal result before consuming the permission-checked private structured-result file. Native Windows is refused; Linux, macOS, and WSL use permission-checked private files. Pi runs from a neutral temporary directory with project resources disabled and `--no-tools`. Kimi (`-p`, `stream-json`) runs from an empty external workspace with a staged `KIMI_CODE_HOME`: sanitized model\u002Fprovider config only (no services, hooks, or extra skill\u002Fagent dirs), its OAuth credential directory linked in and device identity copied so native token refreshes remain durable without exposing the rest of the user's Kimi state. A Markdown `--agent-file` with `tools: []` and `subagents: []` plus an empty `--skills-dir` keep project instructions, tools, and MCP servers out of the review; the prompt travels as the `--prompt` argument, so per-pass prompts are capped at a platform-safe argv budget (120 KiB POSIX, 30 KiB Windows) and larger bundles partition into bounded passes.\n\nAmp cloud\u002Forb agent execution is deliberately unsupported. In current Amp CLI behavior, `--orb-execute` does not preserve the local tool isolation and can expose shell, patch, thread, and reviewer tools. Autoreview therefore never passes `--orb-execute`: the local isolated adapter may call Amp's cloud inference service through `amp.ai.generate`, but it does not launch a cloud Amp agent over an untrusted diff.\n\nCodex uses a named permission profile that grants read access only to an empty temporary workspace. This is narrower than repository-root access, which would expose ignored credentials, and narrower than the legacy `read-only` sandbox, which permits reads across the host filesystem.\n\n## Context Efficiency\n\nRun the helper directly so target selection, engine choice, structured validation, and exit status all stay in one path. If output is noisy, summarize the completed helper output after it returns; do not ask another agent or reviewer to rerun the review.\n\n## Helper\n\nAfter setting `AUTOREVIEW` and `AUTOREVIEW_HARNESS` above:\n\n```bash\n\"$AUTOREVIEW\" --help\n```\n\nThe smoke harness has thin shell wrappers over a shared Python implementation:\n\n```bash\n\"$AUTOREVIEW_HARNESS\" --fixture benign --engine codex\n```\n\nOn native Windows, invoke the extensionless Python helper through Python:\n\n```powershell\npython $AUTOREVIEW --help\n```\n\nand the smoke harness:\n\n```powershell\n& $AUTOREVIEW_HARNESS -Fixture benign -Engine codex\n```\n\nThe helper:\n\n- chooses dirty local changes first\n- accepts `--mode uncommitted` as an alias for `--mode local`\n- otherwise uses current PR base if `gh pr view` works\n- otherwise uses `origin\u002Fmain` for non-main branches\n- does not fetch automatically during branch review; the selected base ref must already resolve locally\n- supports `codex`, `claude`, `amp`, `pi`, and `kimi`; default is `AUTOREVIEW_ENGINE` or `codex`\n- resolves bare `git`, `gh`, reviewer, and PowerShell shell commands from absolute `PATH` entries only, never from the reviewed checkout; explicit `--*-bin` paths are interpreted from the reviewed repository root when relative and accepted only when both the supplied path and resolved target stay outside the reviewed repository\n- use `--mode commit --commit \u003Cref>` for already-committed work, especially clean `main` after landing\n- validates complete Git patches, reviews them in one pass up to the aggregate prompt limit, and automatically uses complete bounded passes above it\n- uses branch mode for committed-only PR work, or explicit local mode with a pinned merge base for a complete PR plus dirty candidate\n- writes reports to stdout and optionally to `--output` or `--json-output` files; preparation progress and provider heartbeats use stderr\n- supports `--dry-run` (validates bundle construction, reviewer CLI resolution, and local isolation startup with version\u002Fhelp probes without contacting a provider; exits nonzero if any check fails), an opt-in per-reviewer wall-clock bound via `--engine-timeout-seconds`, `--prompt`, repo-relative `--prompt-file`, repo-relative `--dataset`, `--no-tools`, `--no-web-search`, repeatable Codex-only safe model\u002Fresponse tuning with `--codex-config key=value`, Codex-only `--codex-speed fast|flex|default`, and commit refs\n- supports `--stream-engine-output` or `AUTOREVIEW_STREAM_ENGINE_OUTPUT=1` for live engine text while preserving structured validation; Codex and Claude hide tool\u002Ffile event details, emit compact activity summaries, and report usage at turn completion\n- supports per-engine `--model`, `--thinking`, and Claude `--fallback-model`\n- uses built-in defaults `codex=gpt-5.6-sol` with `high` reasoning and an access-only `gpt-5.6-terra` retry, `claude=claude-fable-5`, and `amp=openai\u002Fgpt-5.6-sol` with `high` reasoning; honors `AUTOREVIEW_MODEL`, `AUTOREVIEW_THINKING`, `AUTOREVIEW_FALLBACK_MODEL`, and per-engine `AUTOREVIEW_\u003CENGINE>_MODEL` \u002F `AUTOREVIEW_\u003CENGINE>_THINKING` environment overrides when CLI flags are omitted\n- gives Codex the bundle in an empty workspace with web search available; Claude receives the bundle plus WebSearch by default and optional domain-constrained WebFetch; Amp sends the bundle only through direct schema-constrained generation; Pi and Kimi receive the bundle with no tools\n- runs Claude with `--safe-mode` (`v2.1.169+`), `--setting-sources user`, MCP and auto-memory disabled, no filesystem\u002Fshell tools, an empty external workspace, and `--fallback-model` when set\n- runs Amp locally from an empty temporary workspace with isolated runtime roots, complete plugin inventory attestation that fails if any authenticated personal\u002Fworkspace plugin exists, catch-all MCP denial verified by a no-spawn marker probe, a fixed outer trigger, one input-free adapter tool, and direct `amp.ai.generate`; requires `AMP_API_KEY`, refuses native Windows, and refuses cloud\u002Forb agent execution\n- runs Pi `v0.79.0+` from neutral temporary directories with `--no-approve`, `--no-session`, disabled Pi context\u002Fresource loading, and `--no-tools` because its built-in read tools are not repository-confined\n- runs Kimi Code CLI `v0.30.0+` from an empty temporary workspace with a staged `KIMI_CODE_HOME`, sanitized config, an empty `--skills-dir`, and a no-tools\u002Fno-subagents Markdown `--agent-file`\n- prints `review still running: \u003Cengine> elapsed=\u003Cseconds>s pid=\u003Cpid>` to stderr at long-running intervals while waiting for the selected review engine, unless streamed output or compact Codex activity has been visible recently\n- prints `autoreview scoped-clean` only when no findings were rejected or filtered and the provider returned a correct verdict\n- exits nonzero for accepted findings, a provider's incorrect verdict, or an incomplete result\n\n## Result handling\n\nFiltering never rewrites provider correctness, explanation, or confidence.\n`findings` contains accepted findings at the requested priority; local JSON and\ntext results retain `scope_rejected_findings` and `priority_filtered_findings`\nseparately. Any scope rejection makes `review_status` `incomplete` and exits 2,\nincluding when other findings remain or `--expect-findings` is set. Resolve the\ntarget mismatch explicitly; do not infer a broader target from prompt prose.\n\nWith no scope rejection, status is `findings` for accepted findings, `filtered`\nwhen only lower-priority findings remain, `incorrect` for an incorrect verdict\nwithout findings, or `scoped-clean` otherwise. Exit 1 means accepted findings or\nan incorrect provider verdict; even a priority-filtered incorrect verdict stays\nnonzero. Exit 0 for a filtered correct verdict does not certify correctness.\n`--expect-findings` accepts retained findings for harness checks, never scope\nrejections. `--require-finding` checks only accepted findings after both filters\nacross every pass; missing text is retained in `missing_required_findings` and\nexits 2 after writing results.\n\nChunked results retain each provider report under `pass_reports`, print each\nprovider explanation, and use the minimum pass confidence rather than promoting\nconfidence from another pass. Provider JSON remains strictly validated before\nthe helper adds local audit metadata. All engines use this same result path.\n\nThe exact source filename `CredentialFile.swift` is allowed for untracked and\nevidence inputs only outside sensitive parent paths, matching its existing\ntracked-source treatment. This is not a general source-extension exemption:\ncredential stores, credential directories, `.env`, PEM, and key files retain\ntheir exclusions. Reviewers inspect the source content, deleted lines, prompts,\nand datasets present in each outgoing pack for accidentally committed credentials.\n\n## Final Report\n\nReport material findings and the resulting status in chat. If there are none, say so plainly. Do not add command logs, test ledgers, proof blocks, or review receipts unless the user asks for them.\n","# Auto Review\n\nRun the bundled structured review helper only when the user explicitly asks for autoreview, a second-model review, or one of its named review engines. This is code review, not Guardian `auto_review` approval routing.\n\nCodex review is the default when no engine is set. It uses `gpt-5.6-sol` with `high` reasoning by default, then retries once with `gpt-5.6-terra` only when the account cannot access Sol. Claude review is optional and uses `claude-fable-5` by default. Amp review is optional and uses `openai\u002Fgpt-5.6-sol` with `high` reasoning by default. Pi and Kimi use the model configured by their respective CLIs unless `--model` overrides it.\n\nDo not invoke Autoreview automatically before a commit, push, PR, merge, deploy, or final reply. Repository or workflow rules may call it only when they explicitly name it.\n\n## Contract\n\n- Default accepted findings are P0 only: report issues worth blocking the current change\n  because they materially break the normal flow, outcome, or safety boundary.\n  Use `--max-priority P1`, `P2`, or `P3` only when the caller explicitly asks\n  for a wider review.\n- Treat review output as advisory. Never blindly apply it.\n- Verify every finding by reading the real code path and adjacent files.\n- Read dependency docs\u002Fsource\u002Ftypes when the finding depends on external behavior.\n- Reject unrealistic edge cases, speculative risks, unrelated rewrites, and fixes that over-complicate the codebase.\n- Prefer root-cause fixes at the right ownership boundary. A coherent refactor is appropriate when it removes the bug class, duplicate policy, stale paths, or ownership confusion; do not default to a symptom patch.\n- When an accepted finding exposes a bug class or repeated pattern, inspect its owner and relevant sibling implementations before fixing.\n- Fix the same bug class across its owner-boundary neighborhood when practical; stop at unrelated invariants, different owners, and unapproved contract changes.\n- Run one bounded review pass. If an accepted finding changes code, run the smallest relevant test; rerun Autoreview only when the user explicitly requests another pass.\n- For security-audit suppression changes, verify accepted findings remain auditable: suppressed findings stay in structured output, active output keeps an unsuppressible suppression notice, and aggregate findings cannot hide unrelated active risk.\n- Never switch or override the requested review engine\u002Fmodel except for the documented Codex Sol-to-Terra account-access fallback. Capacity, rate-limit, and unrelated failures keep the same engine\u002Fmodel.\n- Be patient with large bundles. Structured review can take up to 30 minutes while the model call is active, especially with Codex tools or web search.\n- Treat heartbeat lines like `review still running: ... elapsed=... pid=...` as healthy progress, not a hang. Let the helper continue while heartbeats are advancing. Pass `--stream-engine-output` when live engine text is useful; Codex and Claude filter tool\u002Ffile chatter, other runnable engines pass raw output through.\n- Do not kill a review just because it has been quiet for 2-5 minutes, or because it is still running under the 30-minute window. Inspect the process only after missing multiple expected heartbeats, after 30 minutes, or after an obviously failed subprocess; prefer letting the same helper command finish.\n- Tools are useful in review mode. Codex receives the validated bundle in an empty workspace so ignored files and linked-worktree metadata remain unreadable; web search stays available for dependency contracts and upstream docs.\n- Security perspective is always included, but it should not cripple legitimate functionality. Report security findings only when the change creates a concrete, actionable risk or removes an important safety check.\n- Reviewer subprocesses preserve engine authentication and non-credentialed proxy variables needed by headless or restricted-network environments while stripping process-injection, Git override, and credentialed proxy values.\n- Credential checking belongs to the reviewer: the hard rules require inspection of the entire change bundle for accidentally committed credentials and require every suspected real credential to be reported as a P0 security finding without reproducing its value. The helper does not invoke an external credential scanner. Security-sensitive paths remain omitted. Safe large diffs are sent as one pass while they fit the aggregate prompt limit, then partitioned into complete bounded passes without truncation.\n- Regression provenance needs patch proof, not blame alone. `git log -S\u002F-G`, `git blame`, commit subjects, and PR metadata locate candidates. Before saying `introduced by`, inspect raw parents with `git --no-replace-objects cat-file -p \u003Csha>` and verify the implicated behavior changed in `git --no-replace-objects diff --no-ext-diff --no-textconv \u003Craw-parent> \u003Csha> -- \u003Cpath>`; a genuine root needs raw-header proof that it has no parents.\n- Blame `^sha`, porcelain `boundary`, and shallow\u002Fgrafted history alone are not introduction proof. `--root` can hide boundary markers; `git show` or `rev-list --parents` can make a shallow boundary look like a root. An available raw parent permits explicit comparison even at a shallow boundary; missing parents or an unverifiable patch require `unknown` with the gap. Use `carried forward` only for verified preexisting behavior and `made visible` only for a verified trigger. Apply the same bar to finding prose, summaries, and owner hints.\n- Keep code author, introducing PR author, merger, committer, automation trigger, and current PR author separate; none of those roles alone proves causation. Cite the verified commit\u002FPR\u002Fdate. If no PR is traceable, use the verified commit and known author identity; unknown identities stay unknown, and missing PR metadata is not a separate finding.\n- For automation merges, identify the human trigger only from explicit timeline\u002Fcomment\u002Fevent evidence, such as a maintainer automerge command or arming label. Report `automerge triggered by @login` only when verified; otherwise say trigger unknown. Triggering or merging is not proof of authorship or introduction.\n- Do not invoke built-in `codex review`, nested reviewers, or review panels from inside the review. The helper builds one validated bundle, calls the selected engine once for normal inputs or once per complete bounded chunk for oversized inputs, validates the structured results, and stops.\n- Stop as soon as the helper exits 0 with no accepted\u002Factionable findings. Do not run an extra review just to get a nicer \"clean\" line, a second opinion, or clearer closeout wording.\n- Treat `scoped-clean` with exit 0 as clean only for the selected Git target and requested priority. `filtered` is not a correctness certificate; `incomplete` requires resolving the scope mismatch or missing required finding before claiming clean.\n- If rejecting a finding as intentional\u002Fnot worth fixing, add a brief inline code comment only when it explains a real invariant or ownership decision that future reviewers should know.\n- If `gh`\u002FGitcrawl reports `database disk image is malformed`, run `gitcrawl doctor --json` once to let the portable cache repair before retrying review; do not bypass the shim unless repair fails and freshness requires live GitHub.\n- If Gitcrawl reports a portable manifest mismatch, source\u002Fruntime DB health error, or stale portable-store checkout, run `gitcrawl doctor --json` and inspect `source_db_health`, `runtime_db_health`, and `portable_store_status` before falling back to live GitHub.\n- Do not push just to review. Push only when the user requested push\u002Fship\u002FPR update.\n\n## Scope\n\nAutoreview does not expand the task. Fix only verified blockers in the requested path. Mention unrelated findings without opening a new workstream, and stop when the requested review pass is complete.\n\n## Skill Path (set once)\n\nSet the skill script paths once, then use `\"$AUTOREVIEW\"` and `\"$AUTOREVIEW_HARNESS\"` in the examples below.\n\nChoose one:\n\n```bash\n# Project-local skill in the current repo for Codex and other agents:\nexport AUTOREVIEW=\".agents\u002Fskills\u002Fautoreview\u002Fscripts\u002Fautoreview\"\nexport AUTOREVIEW_HARNESS=\".agents\u002Fskills\u002Fautoreview\u002Fscripts\u002Ftest-review-harness\"\n```\n\n```bash\n# Claude Code project-local skill in the current repo:\nexport AUTOREVIEW=\".claude\u002Fskills\u002Fautoreview\u002Fscripts\u002Fautoreview\"\nexport AUTOREVIEW_HARNESS=\".claude\u002Fskills\u002Fautoreview\u002Fscripts\u002Ftest-review-harness\"\n```\n\n```bash\n# Source checkout of openclaw\u002Fagent-skills:\nexport AUTOREVIEW=\"skills\u002Fautoreview\u002Fscripts\u002Fautoreview\"\nexport AUTOREVIEW_HARNESS=\"skills\u002Fautoreview\u002Fscripts\u002Ftest-review-harness\"\n```\n\n```bash\n# Global skill:\nexport AGENTS_HOME=\"${AGENTS_HOME:-$HOME\u002F.agents}\"\nexport AUTOREVIEW=\"$AGENTS_HOME\u002Fskills\u002Fautoreview\u002Fscripts\u002Fautoreview\"\nexport AUTOREVIEW_HARNESS=\"$AGENTS_HOME\u002Fskills\u002Fautoreview\u002Fscripts\u002Ftest-review-harness\"\n```\n\nWhen using Claude Code, set `AGENTS_HOME=\"$HOME\u002F.claude\"` for global skills.\n\nOn native Windows, choose the matching pair:\n\n```powershell\n# Project-local skill in the current repo for Codex and other agents:\n$AUTOREVIEW = \".agents\\skills\\autoreview\\scripts\\autoreview\"\n$AUTOREVIEW_HARNESS = \".agents\\skills\\autoreview\\scripts\\test-review-harness.ps1\"\n```\n\n```powershell\n# Claude Code project-local skill in the current repo:\n$AUTOREVIEW = \".claude\\skills\\autoreview\\scripts\\autoreview\"\n$AUTOREVIEW_HARNESS = \".claude\\skills\\autoreview\\scripts\\test-review-harness.ps1\"\n```\n\n```powershell\n# Source checkout of openclaw\u002Fagent-skills:\n$AUTOREVIEW = \"skills\\autoreview\\scripts\\autoreview\"\n$AUTOREVIEW_HARNESS = \"skills\\autoreview\\scripts\\test-review-harness.ps1\"\n```\n\n```powershell\n# Global skill:\n$AgentsHome = if ($env:AGENTS_HOME) { $env:AGENTS_HOME } else { Join-Path $HOME \".agents\" }\n$AUTOREVIEW = Join-Path $AgentsHome \"skills\\autoreview\\scripts\\autoreview\"\n$AUTOREVIEW_HARNESS = Join-Path $AgentsHome \"skills\\autoreview\\scripts\\test-review-harness.ps1\"\n```\n\n## Pick Target\n\nDirty local work relative to HEAD:\n\n```bash\n\"$AUTOREVIEW\" --mode local\n```\n\nWithout `--base`, this reviews HEAD-to-index, index-to-working-tree, and validated\nuntracked files only. `--mode auto` selects the same HEAD-based scope when dirty;\nit does not include the committed PR merely because the checkout is on a PR\nbranch. `--mode uncommitted` is an alias for `--mode local`. A clean local checkout\nwithout an explicit base has no local patch to review.\n\nTo review a dirty candidate against an explicit base, including a resolved merge\nthat has not been committed:\n\n```bash\n\"$AUTOREVIEW\" --mode local --base origin\u002Fmain\n```\n\nThe helper pins that base to a commit at target selection. It reviews base-to-index\nchanges and index-to-working-tree changes separately, plus validated untracked\nfiles. Only files identical across the base, index, and working tree are outside\nthe change bundle; staged changes later undone remain included. Actual binary\nor submodule changes still refuse review. The bundle labels its pinned\nstaged base. Git status remains relative to HEAD and does not define review scope.\nWithout `--base`, local mode retains its usual HEAD-to-index behavior; it never\ninfers a base from an in-progress merge.\n\nCommitted-only branch\u002FPR work:\n\n```bash\n\"$AUTOREVIEW\" --mode branch --base origin\u002Fmain\n```\n\nBranch mode reviews `BASE...HEAD` (merge-base-to-HEAD); staged, unstaged, and\nuntracked changes are excluded even in a dirty checkout. To review the **complete\nPR candidate including dirty rewrites**, pin the PR merge base and use local mode:\n\n```bash\npr_base=$(gh pr view --json baseRefName --jq .baseRefName)\nmerge_base=$(git merge-base HEAD \"origin\u002F$pr_base\")\n\"$AUTOREVIEW\" --mode local --base \"$merge_base\"\n```\n\nThe remote base must already be available and current locally; the helper does\nnot fetch. The pinned merge base avoids including unrelated upstream changes.\nAdd `--max-priority P2` when the caller requests P2 findings. An explicit `--base`\nalso applies when auto selects local, but explicit local mode avoids changing\ntargets when the checkout becomes clean.\n\nOptional review context is first-class. Prompt files and datasets must be repo-relative so review bundles cannot pull arbitrary host files. Context never expands the selected Git target, even if it contains a complete candidate diff or asks to review the whole PR. Finding membership is checked by changed file, not individual hunk:\n\n```bash\n\"$AUTOREVIEW\" --mode branch --base origin\u002Fmain --prompt-file review-notes.md --dataset evidence.json\n```\n\nIf an open PR exists, use its actual base:\n\n```bash\nbase=$(gh pr view --json baseRefName --jq .baseRefName)\n\"$AUTOREVIEW\" --mode branch --base \"origin\u002F$base\"\n```\n\nCommitted single change:\n\n```bash\n\"$AUTOREVIEW\" --mode commit --commit HEAD\n```\n\nUse commit review for already-landed or already-pushed work on `main`. Reviewing\nclean `main` against `origin\u002Fmain` is usually an empty diff after push. For a\nsmall stack, review each commit explicitly or review the branch before merging\nwith `--base`. Commit review compares the raw recorded parent with the selected\ncommit, ignoring replacement refs and legacy grafts. A genuine root compares\nagainst the empty tree. Missing parent objects stop review: explicitly deepen\u002Ffetch the needed\nhistory and rerun. The helper does not fetch it automatically.\n\n## Oversized Bundles\n\nThe helper validates the complete patch before partitioning it. A bundle that fits\nthe aggregate prompt limit remains one integrated review pass. Larger bundles\nare split at bundle sections and file boundaries where possible; an oversized\nsingle-file block is split at line boundaries with repeated file\u002Fhunk context\nand an absolute new- or old-file line offset. Untracked snapshots use\ninjection-safe source-line records so continuation passes retain reportable\nlocations. A single physical diff line split across passes also retains its\noriginal addition, deletion, or context marker.\nPrompt instructions remain whole in every pass. Large datasets are grouped from\ntheir validated file records, never by reparsing headings inside evidence.\nIndividual oversized datasets split at lines or UTF-8 boundaries with their\noriginal path and byte offset. Each evidence batch is paired with the complete\nchange bundle: every original change byte appears exactly once per evidence\nbatch, and every evidence byte is retained. All validated reports are merged\nbefore required-finding and exit-status checks.\nThere is no fixed pass-count ceiling: the complete frozen input determines the\nfinite pass sequence. The helper prints its size and pass count before running\npasses serially. Each pass retains the same prompt-size limit and\nreviewer isolation; a failed pass aborts without publishing a partial verdict.\n\nPreparation prints immediate phase updates and periodic elapsed time to stderr,\nwith file\u002Fbyte counts during hashing and no filenames or contents. These updates\nare separate from provider heartbeats and do not count against engine deadlines.\nBundle construction captures finding membership alongside validated text; normal\nand dry runs reuse that record without reopening untracked files for membership.\n\nDry runs reuse captured inputs and prompt construction without whole-tree integrity sweeps. Real\nreviews retain full fresh tree hashing before bundle construction, before review,\nand before publication, including unrelated tracked files, nonignored untracked files, index\nstate, and initialized submodules. Explicit prompt files and datasets also retain\ntheir own frozen bytes and raw path identities, regardless of Git ignore status\nor finding scope. They are revalidated before sending each pass and before\npublication; content changes, replacements, and leaf or ancestor symlink swaps\nrefuse stale results. These endpoint checks are not atomic filesystem snapshots\nand cannot guarantee detection of transient changes restored between checks.\n\nEvidence batches can multiply the pass count. Chunking cannot give one model\ncall every cross-file implementation detail. For architecture-heavy changes, still prefer\na coherent branch or PR shape whose semantic decision surface fits one pass.\nRemoving verified non-authoritative generated noise remains useful, but never\ndrop lockfiles, generated clients, policies, manifests, schemas, or other\nindependently semantic artifacts merely to shrink the review.\n\n## Models and thinking\n\nThe helper accepts `--model` globally or per engine (`engine=model`) and `--thinking` globally or per engine (`engine=level`). Repeat either flag for multiple reviewers.\n\nRecommended model defaults:\n\n| Engine              | Default model                                      | Source note                                           |\n| ------------------- | -------------------------------------------------- | ----------------------------------------------------- |\n| **codex** (default) | `gpt-5.6-sol` -> `gpt-5.6-terra` on access failure | OpenClaw org review default                           |\n| **claude**          | `claude-fable-5`                                   | Anthropic's most capable widely released Claude model |\n| **amp**             | `openai\u002Fgpt-5.6-sol`                               | Amp structured-generation review default              |\n\nCLI flags and environment variables override these defaults. Amp model IDs must use `provider\u002Fmodel` form. Pi and Kimi do not get built-in model defaults because their configured model catalogs may vary by installation.\n\n| Engine              | Model flag                 | Example model IDs                                                            | Thinking flag                             | Accepted levels                                            |\n| ------------------- | -------------------------- | ---------------------------------------------------------------------------- | ----------------------------------------- | ---------------------------------------------------------- |\n| **codex** (default) | `codex --model X exec ...` | `gpt-5.6-sol`, then `gpt-5.6-terra` on Sol access failure                    | `-c model_reasoning_effort=Y`             | `none`, `minimal`, `low`, `medium`, `high`, `xhigh`, `max` |\n| **claude**          | `claude --model X`         | `claude-fable-5`, `claude-opus-4-8`, `claude-sonnet-4-6`, `claude-haiku-4-5` | `--effort Y`                              | `low`, `medium`, `high`, `xhigh`, `max`                    |\n| **amp**             | Amp `amp.ai.generate`      | `openai\u002Fgpt-5.6-sol`                                                         | `reasoningEffort`                         | `none`, `low`, `medium`, `high`, `xhigh`, `max`            |\n| **pi**              | `pi --model X`             | `anthropic\u002Fclaude-sonnet-4`, `openai\u002Fgpt-4o`                                 | `--thinking Y`                            | `off`, `minimal`, `low`, `medium`, `high`, `xhigh`         |\n| **kimi**            | `kimi --model X`           | A model alias from the user's Kimi config                                    | `[thinking] enabled` in the staged config | `on`, `off`                                                |\n\nClaude also supports `--fallback-model a,b` for availability-based fallback chains ([model-config](https:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fmodel-config)). Current Claude docs note that auth, billing, rate-limit, request-size, and transport errors do not trigger fallback, and the changelog documents interactive-session support in `v2.1.166`.\n\n[OpenAI's model guidance](https:\u002F\u002Fdevelopers.openai.com\u002Fapi\u002Fdocs\u002Fguides\u002Flatest-model) identifies Sol as the GPT-5.6 frontier-capability route and documents `max` support. Autoreview keeps `high` as its default; use `max` only for the hardest quality-first reviews after comparing its latency and cost with `xhigh` on representative changes.\n\nExamples matching current `main` behavior:\n\n```bash\n# Codex with explicit model and reasoning\n\"$AUTOREVIEW\" --engine codex --model gpt-5.6-sol --thinking high\n\n# Codex fast mode (priority service tier); needs a model whose catalog lists the tier, silently standard otherwise\n\"$AUTOREVIEW\" --engine codex --codex-speed fast\n\n# Safe Codex model\u002Fresponse tuning overrides (--codex-speed wins over a service_tier here)\n\"$AUTOREVIEW\" --engine codex --codex-config 'service_tier=\"fast\"'\n\n# Claude Code aliases or full model names, with optional availability fallback\n\"$AUTOREVIEW\" --engine claude --model claude-fable-5 --thinking max\n\"$AUTOREVIEW\" --engine claude --model claude-fable-5 --fallback-model claude-opus-4-8,claude-sonnet-4-6\n\n# Amp direct structured generation (requires AMP_API_KEY)\n\"$AUTOREVIEW\" --engine amp --model openai\u002Fgpt-5.6-sol --thinking high --amp-bin amp\n\n# Pi with explicit model and thinking level\n\"$AUTOREVIEW\" --engine pi --model anthropic\u002Fclaude-sonnet-4 --thinking high --pi-bin pi\n\n# Kimi with its configured default model, or a configured model alias\n\"$AUTOREVIEW\" --engine kimi --thinking on --kimi-bin kimi\n\"$AUTOREVIEW\" --engine kimi --model kimi-model-alias\n\n```\n\n### Environment defaults\n\nCLI flags take precedence over environment variables.\n\nStore persistent personal defaults in your shell startup file or launcher\nenvironment. For repository-local defaults, use an existing local environment\nloader such as an untracked `.envrc`; the helper does not write a config file.\n\n| Variable                            | Purpose                                                                                                                          |\n| ----------------------------------- | -------------------------------------------------------------------------------------------------------------------------------- |\n| `AUTOREVIEW_MODEL`                  | Override the built-in default `--model` for all engines                                                                          |\n| `AUTOREVIEW_THINKING`               | Default `--thinking` for all engines                                                                                             |\n| `AUTOREVIEW_FALLBACK_MODEL`         | Default Claude `--fallback-model` chain                                                                                          |\n| `AUTOREVIEW_ENGINE_TIMEOUT_SECONDS` | Optional positive wall-clock limit for each reviewer process; disabled by default                                                |\n| `AUTOREVIEW_\u003CENGINE>_MODEL`         | Per-engine model override, for example `AUTOREVIEW_CODEX_MODEL=gpt-5.6-sol`                                                      |\n| `AUTOREVIEW_\u003CENGINE>_THINKING`      | Per-engine thinking override                                                                                                     |\n| `AUTOREVIEW_CODEX_CONFIG`           | Safe Codex model\u002Fresponse tuning overrides, semicolon-separated, e.g. `service_tier=\"fast\"`; capability-bearing keys fail closed |\n| `AUTOREVIEW_CODEX_SPEED`            | Codex service tier override: `fast` (priority), `flex`, or `default`; silently standard when the model does not list the tier    |\n| `AUTOREVIEW_CLAUDE_FALLBACK_MODEL`  | Claude-only fallback chain                                                                                                       |\n| `AUTOREVIEW_PROVIDER_ENV_ALLOW`     | Comma-separated custom Pi credential variable names; names must end in a recognized credential suffix                            |\n| `AMP_API_KEY`                       | Required Amp API credential; file\u002Fkeychain auth is intentionally excluded from the isolated runtime                              |\n\nCodex maps thinking to `model_reasoning_effort`. Claude maps thinking to `--effort`. Amp maps thinking to `amp.ai.generate.reasoningEffort`. Pi maps thinking to `--thinking`. Kimi maps `on` and `off` to `[thinking] enabled` in the staged review config. Only Claude accepts `--fallback-model`; global CLI\u002Fenv fallback requires at least one Claude reviewer, and engine-specific fallback overrides require that reviewer to be selected. Non-Claude fallback overrides, including `AUTOREVIEW_\u003CNONCLAUDE>_FALLBACK_MODEL`, fail closed instead of being silently ignored.\n\nAmp receives only `AMP_API_KEY` from the caller. Autoreview intentionally ignores `AMP_URL`, user settings, stored authentication, inherited MCP configuration, and other runtime variables. The API key's authenticated account and workspace must have no personal or workspace plugins: current normal Amp execution loads every authenticated plugin, so the preflight requests the complete inventory and fails before creating the review prompt unless the generated adapter is the only plugin. A dedicated Amp API key\u002Faccount without plugins is the safest setup. Amp can still discover personal and workspace skill metadata, but the isolated settings deny every local and remote MCP server before use, and the outer adapter has no skill tool. Custom Amp endpoints are not supported because forwarding an arbitrary endpoint could disclose the API key and review bundle. Native Windows is refused because its `chmod` behavior cannot establish or attest the POSIX private-file permissions used here; use Linux, macOS, or WSL.\n\n## Review engine isolation\n\nWhen autoreview runs inside the repository under review, external reviewer CLIs must not load project-local trust or configuration that the branch controls.\n\n| Engine     | Isolation flags                                                                                                                                                                                                                                                                                               | Reference                                                                      |\n| ---------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------ |\n| **codex**  | Auth-only config overrides, isolated workspace, `exec --ignore-user-config --ignore-rules --skip-git-repo-check`, plus read-only sandbox                                                                                                                                                                      | Codex CLI `exec --help`                                                        |\n| **claude** | `--safe-mode --setting-sources user --strict-mcp-config --disallowedTools mcp__*`; auto-memory and filesystem\u002Fshell tools disabled; empty external workspace; WebSearch by default (`v2.1.169+`)                                                                                                              | Claude Code [CLI reference](https:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fcli-reference)     |\n| **amp**    | Empty external workspace and isolated HOME\u002FXDG roots; complete authenticated plugin inventory must contain only the generated adapter; catch-all MCP denial with a process-spawn probe; fixed outer trigger and one input-free adapter tool; private prompt only reaches schema-constrained `amp.ai.generate` | Amp [plugin API](https:\u002F\u002Fampcode.com\u002Fmanual\u002Fplugin-api) and local CLI `--help` |\n| **pi**     | `--no-approve --no-session --no-context-files --no-extensions --no-skills --no-prompt-templates --no-themes --no-tools`                                                                                                                                                                                       | Pi CLI `--help`; requires Pi `v0.79.0+`                                        |\n| **kimi**   | Empty external workspace; staged `KIMI_CODE_HOME` with sanitized config; Markdown custom agent with no tools\u002Fsubagents; explicit empty `--skills-dir`; isolated runtime state                                                                                                                                 | Kimi Code CLI `--help`; requires Kimi `v0.30.0+`                               |\n\nCodex `--ignore-user-config` skips config loading for the exec run. Autoreview reconstructs only the documented `cli_auth_credentials_store`, `forced_login_method`, and `forced_chatgpt_workspace_id` settings from `CODEX_HOME\u002Fconfig.toml`, keeping authentication usable without forwarding unrelated user configuration. Codex runs in an empty temporary workspace: the validated bundle is its sole repository input, ignored files and linked-worktree metadata remain unreadable, and the zero project-doc budget keeps workspace instructions out of the prompt. `--ignore-rules` skips user\u002Fproject execpolicy rules. Claude `--safe-mode` disables project hooks, skills, plugins, MCP servers, and CLAUDE.md; autoreview supplies WebSearch by default, permits only explicitly domain-constrained WebFetch rules, and exposes no filesystem or shell tools. Amp runs its local CLI with isolated HOME\u002FXDG roots and one generated adapter plugin whose name includes a fresh 128-bit random suffix. Current normal Amp execution loads all authenticated plugins, so the preflight deliberately requests that same complete inventory and fails before writing the private prompt unless the generated adapter is the only active plugin, with exactly its expected tool, agent, and mode. Users with personal or workspace plugins must use a dedicated plugin-free Amp API key\u002Faccount. Isolated `amp.mcpPermissions` reject every local command and remote URL. Before writing the private prompt, autoreview creates a temporary skill whose MCP command would write a marker, runs `amp tools list`, and requires Amp to report the policy rejection without creating the marker; it removes that skill before continuing. A custom outer mode then receives only a fixed harmless trigger and exposes exactly one trusted, input-free `autoreview_generate` tool. That tool reads the private prompt file and calls `amp.ai.generate` directly with an explicit system prompt and report schema, so the untrusted patch never enters the outer agent context. Autoreview requires the stream's leading init event to attest the empty working directory, the exact singleton adapter-tool inventory, and `mcp_servers: []`; it then requires exactly one correctly ordered empty-input tool call\u002Fresult and one terminal result before consuming the permission-checked private structured-result file. Native Windows is refused; Linux, macOS, and WSL use permission-checked private files. Pi runs from a neutral temporary directory with project resources disabled and `--no-tools`. Kimi (`-p`, `stream-json`) runs from an empty external workspace with a staged `KIMI_CODE_HOME`: sanitized model\u002Fprovider config only (no services, hooks, or extra skill\u002Fagent dirs), its OAuth credential directory linked in and device identity copied so native token refreshes remain durable without exposing the rest of the user's Kimi state. A Markdown `--agent-file` with `tools: []` and `subagents: []` plus an empty `--skills-dir` keep project instructions, tools, and MCP servers out of the review; the prompt travels as the `--prompt` argument, so per-pass prompts are capped at a platform-safe argv budget (120 KiB POSIX, 30 KiB Windows) and larger bundles partition into bounded passes.\n\nAmp cloud\u002Forb agent execution is deliberately unsupported. In current Amp CLI behavior, `--orb-execute` does not preserve the local tool isolation and can expose shell, patch, thread, and reviewer tools. Autoreview therefore never passes `--orb-execute`: the local isolated adapter may call Amp's cloud inference service through `amp.ai.generate`, but it does not launch a cloud Amp agent over an untrusted diff.\n\nCodex uses a named permission profile that grants read access only to an empty temporary workspace. This is narrower than repository-root access, which would expose ignored credentials, and narrower than the legacy `read-only` sandbox, which permits reads across the host filesystem.\n\n## Context Efficiency\n\nRun the helper directly so target selection, engine choice, structured validation, and exit status all stay in one path. If output is noisy, summarize the completed helper output after it returns; do not ask another agent or reviewer to rerun the review.\n\n## Helper\n\nAfter setting `AUTOREVIEW` and `AUTOREVIEW_HARNESS` above:\n\n```bash\n\"$AUTOREVIEW\" --help\n```\n\nThe smoke harness has thin shell wrappers over a shared Python implementation:\n\n```bash\n\"$AUTOREVIEW_HARNESS\" --fixture benign --engine codex\n```\n\nOn native Windows, invoke the extensionless Python helper through Python:\n\n```powershell\npython $AUTOREVIEW --help\n```\n\nand the smoke harness:\n\n```powershell\n& $AUTOREVIEW_HARNESS -Fixture benign -Engine codex\n```\n\nThe helper:\n\n- chooses dirty local changes first\n- accepts `--mode uncommitted` as an alias for `--mode local`\n- otherwise uses current PR base if `gh pr view` works\n- otherwise uses `origin\u002Fmain` for non-main branches\n- does not fetch automatically during branch review; the selected base ref must already resolve locally\n- supports `codex`, `claude`, `amp`, `pi`, and `kimi`; default is `AUTOREVIEW_ENGINE` or `codex`\n- resolves bare `git`, `gh`, reviewer, and PowerShell shell commands from absolute `PATH` entries only, never from the reviewed checkout; explicit `--*-bin` paths are interpreted from the reviewed repository root when relative and accepted only when both the supplied path and resolved target stay outside the reviewed repository\n- use `--mode commit --commit \u003Cref>` for already-committed work, especially clean `main` after landing\n- validates complete Git patches, reviews them in one pass up to the aggregate prompt limit, and automatically uses complete bounded passes above it\n- uses branch mode for committed-only PR work, or explicit local mode with a pinned merge base for a complete PR plus dirty candidate\n- writes reports to stdout and optionally to `--output` or `--json-output` files; preparation progress and provider heartbeats use stderr\n- supports `--dry-run` (validates bundle construction, reviewer CLI resolution, and local isolation startup with version\u002Fhelp probes without contacting a provider; exits nonzero if any check fails), an opt-in per-reviewer wall-clock bound via `--engine-timeout-seconds`, `--prompt`, repo-relative `--prompt-file`, repo-relative `--dataset`, `--no-tools`, `--no-web-search`, repeatable Codex-only safe model\u002Fresponse tuning with `--codex-config key=value`, Codex-only `--codex-speed fast|flex|default`, and commit refs\n- supports `--stream-engine-output` or `AUTOREVIEW_STREAM_ENGINE_OUTPUT=1` for live engine text while preserving structured validation; Codex and Claude hide tool\u002Ffile event details, emit compact activity summaries, and report usage at turn completion\n- supports per-engine `--model`, `--thinking`, and Claude `--fallback-model`\n- uses built-in defaults `codex=gpt-5.6-sol` with `high` reasoning and an access-only `gpt-5.6-terra` retry, `claude=claude-fable-5`, and `amp=openai\u002Fgpt-5.6-sol` with `high` reasoning; honors `AUTOREVIEW_MODEL`, `AUTOREVIEW_THINKING`, `AUTOREVIEW_FALLBACK_MODEL`, and per-engine `AUTOREVIEW_\u003CENGINE>_MODEL` \u002F `AUTOREVIEW_\u003CENGINE>_THINKING` environment overrides when CLI flags are omitted\n- gives Codex the bundle in an empty workspace with web search available; Claude receives the bundle plus WebSearch by default and optional domain-constrained WebFetch; Amp sends the bundle only through direct schema-constrained generation; Pi and Kimi receive the bundle with no tools\n- runs Claude with `--safe-mode` (`v2.1.169+`), `--setting-sources user`, MCP and auto-memory disabled, no filesystem\u002Fshell tools, an empty external workspace, and `--fallback-model` when set\n- runs Amp locally from an empty temporary workspace with isolated runtime roots, complete plugin inventory attestation that fails if any authenticated personal\u002Fworkspace plugin exists, catch-all MCP denial verified by a no-spawn marker probe, a fixed outer trigger, one input-free adapter tool, and direct `amp.ai.generate`; requires `AMP_API_KEY`, refuses native Windows, and refuses cloud\u002Forb agent execution\n- runs Pi `v0.79.0+` from neutral temporary directories with `--no-approve`, `--no-session`, disabled Pi context\u002Fresource loading, and `--no-tools` because its built-in read tools are not repository-confined\n- runs Kimi Code CLI `v0.30.0+` from an empty temporary workspace with a staged `KIMI_CODE_HOME`, sanitized config, an empty `--skills-dir`, and a no-tools\u002Fno-subagents Markdown `--agent-file`\n- prints `review still running: \u003Cengine> elapsed=\u003Cseconds>s pid=\u003Cpid>` to stderr at long-running intervals while waiting for the selected review engine, unless streamed output or compact Codex activity has been visible recently\n- prints `autoreview scoped-clean` only when no findings were rejected or filtered and the provider returned a correct verdict\n- exits nonzero for accepted findings, a provider's incorrect verdict, or an incomplete result\n\n## Result handling\n\nFiltering never rewrites provider correctness, explanation, or confidence.\n`findings` contains accepted findings at the requested priority; local JSON and\ntext results retain `scope_rejected_findings` and `priority_filtered_findings`\nseparately. Any scope rejection makes `review_status` `incomplete` and exits 2,\nincluding when other findings remain or `--expect-findings` is set. Resolve the\ntarget mismatch explicitly; do not infer a broader target from prompt prose.\n\nWith no scope rejection, status is `findings` for accepted findings, `filtered`\nwhen only lower-priority findings remain, `incorrect` for an incorrect verdict\nwithout findings, or `scoped-clean` otherwise. Exit 1 means accepted findings or\nan incorrect provider verdict; even a priority-filtered incorrect verdict stays\nnonzero. Exit 0 for a filtered correct verdict does not certify correctness.\n`--expect-findings` accepts retained findings for harness checks, never scope\nrejections. `--require-finding` checks only accepted findings after both filters\nacross every pass; missing text is retained in `missing_required_findings` and\nexits 2 after writing results.\n\nChunked results retain each provider report under `pass_reports`, print each\nprovider explanation, and use the minimum pass confidence rather than promoting\nconfidence from another pass. Provider JSON remains strictly validated before\nthe helper adds local audit metadata. All engines use this same result path.\n\nThe exact source filename `CredentialFile.swift` is allowed for untracked and\nevidence inputs only outside sensitive parent paths, matching its existing\ntracked-source treatment. This is not a general source-extension exemption:\ncredential stores, credential directories, `.env`, PEM, and key files retain\ntheir exclusions. Reviewers inspect the source content, deleted lines, prompts,\nand datasets present in each outgoing pack for accidentally committed credentials.\n\n## Final Report\n\nReport material findings and the resulting status in chat. If there are none, say so plainly. Do not add command logs, test ledgers, proof blocks, or review receipts unless the user asks for them.\n",{"name":5,"description":7},"main","d69096e239f2a89aabc2cd078c47664a9d3b1e11","07-后端与架构\u002Fautoreview",3,"https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts","https:\u002F\u002Fgithub.com\u002Flza6","个人vx：Tf00798\r\n个人✈：@TFGY999","2026-10-08 14:29:22",{"source":8,"skillName":5,"commands":33,"prompt":43,"zipAvailable":44},[34,37,40],{"tool":35,"command":36},"npx","npx skills add lza6\u002FArmor-piercing-system-prompts --skill autoreview",{"tool":38,"command":39},"bunx","bunx skills add lza6\u002FArmor-piercing-system-prompts --skill autoreview",{"tool":41,"command":42},"pnpm","pnpm dlx skills add lza6\u002FArmor-piercing-system-prompts --skill autoreview","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fautoreview\u002FSKILL.md",true,[46,52],{"path":47,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":50,"sourceHtmlUrl":51},"AGENTS.md","REFERENCE","参考文档",437,"https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fautoreview\u002FAGENTS.md",{"path":53,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":50,"sourceHtmlUrl":54},"CLAUDE.md","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fautoreview\u002FCLAUDE.md",[56,63,68,75,82,89],{"skillNo":57,"name":58,"nameZh":6,"description":59,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":60,"categories":61,"keywords":62,"publishedAt":19},"01M7FFG485KY83TD6HSC4YPCMC","obsidian-clipper-template-creator","Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fobsidian-clipper-template-creator\u002FSKILL.md",[],[],{"skillNo":64,"name":58,"nameZh":6,"description":59,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":65,"categories":66,"keywords":67,"publishedAt":19},"01M7FFG0T6AZ0STEJPSB3CX84K","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fobsidian-clipper-template-creator-antigravity-awesome-skills-main\u002FSKILL.md",[],[],{"skillNo":69,"name":70,"nameZh":6,"description":71,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":72,"categories":73,"keywords":74,"publishedAt":19},"01M7FFFY0QEM14SKWPDYGMXM2P","python-api","Programmatic Object-Detection-Metrics API guidance for BoundingBox, BoundingBoxes, Evaluator, enums, coordinate conversion, AP\u002FmAP semantics, and source-style import diagnostics.","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fobject-detection-metrics\u002Fsub-skills\u002Fpython-api\u002FSKILL.md",[],[],{"skillNo":76,"name":77,"nameZh":6,"description":78,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":79,"categories":80,"keywords":81,"publishedAt":19},"01M7FFFVZQZ2GJ0C0TGP7JHD20","file-evaluation","Folder-based PASCAL VOC AP\u002FmAP evaluation from ground-truth and detection text files.","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fobject-detection-metrics\u002Fsub-skills\u002Ffile-evaluation\u002FSKILL.md",[],[],{"skillNo":83,"name":84,"nameZh":6,"description":85,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":86,"categories":87,"keywords":88,"publishedAt":19},"01M7FFFSFNYSE77SJ6PE1CK2DA","object-detection-metrics","Guides legacy Object-Detection-Metrics PASCAL VOC AP\u002FmAP evaluation from detection text folders and direct Python API classes.","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fobject-detection-metrics\u002FSKILL.md",[],[],{"skillNo":90,"name":91,"nameZh":6,"description":92,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":93,"categories":94,"keywords":95,"publishedAt":19},"01M7FFFP0GCK1RVXBCPQ8APBMT","nx-workspace-patterns","Configure and optimize Nx monorepo workspaces. Use when setting up Nx, configuring project boundaries, optimizing build caching, or implementing affected commands.","https:\u002F\u002Fgithub.com\u002Flza6\u002FArmor-piercing-system-prompts\u002Fblob\u002Fmain\u002F07-后端与架构\u002Fnx-workspace-patterns\u002FSKILL.md",[],[]]