[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M782MMY01GKYGJPYMF1VECCB":3,"skill-same:01M782MMY01GKYGJPYMF1VECCB":47},{"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":25,"commitSha":26,"skillDirectory":27,"forksCount":28,"repositoryHtmlUrl":29,"repositoryHomepageUrl":6,"ownerHtmlUrl":30,"ownerBlogUrl":31,"ownerDescription":32,"updatedAt":33,"install":34,"resources":46},"01M782MMY01GKYGJPYMF1VECCB","autopilot",null,"Run one explicit bounded Blueprint spec and implementation pass through configured quality gates, then stop before completion or external actions. Use only when the user directly invokes \u002Fautopilot, $autopilot, or asks for Autopilot.","bradtraversy\u002Fskillpass","bradtraversy","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F5550850?v=4","01M782H8BT6A97A17XDFHDABWZ","Brad Traversy","01M782H8CDJZW3HYDHX23FA9VF",26,"MIT","https:\u002F\u002Fgithub.com\u002Fbradtraversy\u002Fskillpass\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fautopilot\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: autopilot\ndescription: Run one explicit bounded Blueprint spec and implementation pass through configured quality gates, then stop before completion or external actions. Use only when the user directly invokes \u002Fautopilot, $autopilot, or asks for Autopilot.\ndisable-model-invocation: true\n---\n\n# autopilot - optional Blueprint loop\n\n**Context reuse:** Reuse any required file already loaded in project instructions or the current session. Read it again only if absent, changed, or exact current bytes or line references are needed.\n\n**First action:** Before project inspection, preflight, or any other tool call,\npublish `running` to `blueprint\u002F.state\u002Frun.json` using the dashboard activity\ncontract in `AGENTS.md`.\n\nWhere this sits in the workflow:\n\n    \u002Fstatus  ->  [autopilot]  ->  review packet  ->  \u002Fcomplete\n    (where       (spec, build,      (human review,    (log, commit,\n     are we?)     configured gates)  fixes if needed)  merge with approval)\n\nAutopilot is an explicit opt-in path. It uses the same Blueprint files and the\nsame quality gates, but it does not stop after every normal review point. A\nsingle user request is permission to run one bounded loop until the feature is\nready for review, blocked, or unsafe to continue.\n\nIt combines `\u002Ffeature` or `\u002Ffix` with `\u002Fimplement` and continues through the\nspec-review stop retained by the normal workflow. That human spec approval is\nthe main control Autopilot intentionally removes. It does not remove the final\nreview packet or the option to walk through the completed code.\n\nIt does **not** replace the normal workflow. `\u002Ffeature`, `\u002Fimplement`, `\u002Fcheck`,\nand `\u002Fcomplete` remain the conservative default.\n\nDo not suggest Autopilot as the default next action. Use it only when the user\nexplicitly asks for it.\n\nThe explicit Autopilot request is permission to create checkpoint commits on the\nfeature or fix branch after passing implementation steps when\n`workflow.checkpointCommits` is enabled. It is not permission to merge, push,\ndeploy, publish, send, delete data, or run destructive actions.\n\n## Input\n\nCommon forms:\n\n- No argument: resume the current feature if one exists, otherwise target the next\n  unchecked build-plan item.\n- A number or name: target that build-plan feature, for example `\u002Fautopilot 3` or\n  `$autopilot \"directory listing\"`.\n- `fix \"\u003Cissue>\"`: write and build an ad-hoc fix spec.\n- `resume`: continue the current feature on its existing branch.\n\nIf the requested target conflicts with a feature already in progress, stop and\nask which one should win. Do not overwrite `blueprint\u002Fcontext\u002Fcurrent-feature.md`\nsilently.\n\nRollback is intentionally excluded from Autopilot. If the request is a rollback\nor `current-feature.md` is marked `Type: Rollback`, stop and direct the user to\nthe reviewed `\u002Fimplement` path. Reversing completed work requires the explicit\ndependency and conflict gates in `\u002Frollback` and `\u002Fimplement`.\n\n## Step 1 - preflight like \u002Fstatus\n\nRead the same state `\u002Fstatus` reads:\n\n- `AGENTS.md`\n- `blueprint\u002Fconfig.json`\n- `blueprint\u002Fproject-plan.md`\n- `blueprint\u002Fbuild-plan.md`\n- `blueprint\u002Fcontext\u002Fproject-overview.md`\n- `blueprint\u002Fcontext\u002Fcurrent-feature.md`\n- `blueprint\u002Fcontext\u002Ffindings.md`\n- `blueprint\u002Fcontext\u002Fcoding-standards.md`\n- `blueprint\u002Fcontext\u002Fai-interaction.md`\n- git branch, status, and recent log\n\nThen decide whether it is safe to run.\n\nStop before changing files when:\n\n- `blueprint\u002Fconfig.json` exists but is invalid. Point the user to `\u002Fdoctor`.\n- The repo is not a git repo.\n- The working tree is dirty and there is no current feature tying those changes\n  to this run.\n- `current-feature.md` has real work and the user requested a different target.\n- `project-overview.md` is missing or stale and the planning docs are not clear\n  enough to regenerate it.\n- The next feature is visual or replication-heavy and no design reference exists.\n- The task needs product, data, auth, billing, or destructive decisions the docs\n  do not answer.\n\nIf the only issue is that `project-overview.md` is stale and the plans are clear,\nregenerate it using the `\u002Foverview` behavior and continue. Include that in the\nfinal packet.\n\nThe initial `blueprint\u002F.state\u002Frun.json` record required by `AGENTS.md` must\nalready show command `autopilot` and status `running` before preflight begins.\nAfter preflight passes, enrich it with boundary `reviewed`, the target feature\nor fix, and build-step progress when known. Update it after the spec, each\npassing build step, and each configured gate. On a hard\nstop, set status `blocked` with `\u002Fautopilot resume` when resuming is safe. At the\nfinal review packet, set status `ready` because Autopilot stops before\n`\u002Fcomplete`. Activity reporting must never weaken or block the workflow itself.\n\n## Step 2 - choose or write the spec\n\nIf `blueprint\u002Fcontext\u002Fcurrent-feature.md` already contains an active spec,\nresume it. Read checked steps and continue from the first unchecked step.\n\nIf there is no active spec:\n\n1. Use the `\u002Ffeature` behavior for a planned feature, or `\u002Ffix` behavior for a\n   requested fix.\n2. Write `blueprint\u002Fcontext\u002Fcurrent-feature.md`.\n3. Red-team the spec before building:\n   - missing unhappy paths\n   - oversized steps\n   - undefined contracts\n   - missing design reference\n   - scope creep\n   - vague done-whens\n   - missing testing plan when `AGENTS.md` declares a test command\n4. Apply the spec fixes.\n\nAutopilot may continue past this spec gate because the user explicitly invoked\nAutopilot. Still report what the critique changed in the final packet.\nFollow the proportional-engineering contract in `AGENTS.md` throughout this run.\n\n## Step 3 - create or reuse the branch\n\nUse the same branch rules as `\u002Fimplement`:\n\n- Feature: the configured feature prefix, default `feature\u002F\u003Cname>`\n- Fix: the configured fix prefix, default `fix\u002F\u003Cname>`\n\nIf the branch already exists, switch to it only if it matches the active spec.\nIf switching branches would strand unrelated dirty work, stop and report the\nproblem.\n\n## Step 4 - implement in small steps\n\nWork through the spec's build steps in order. Each step must remain reviewable.\nDo not pause for user approval after each passing step, regardless of the\nconfigured `workflow.stepReview` value. The review happens at the final packet\nunless a hard stop is hit.\n\nFor every step:\n\n1. Implement only that step.\n2. Run the relevant verification:\n   - the exact `Verify` command from `AGENTS.md`, when declared\n   - otherwise the build, relevant test, lint, and typecheck commands already\n     documented by the project\n   - browser, CLI, API, or app-level evidence for behavioral done-whens\n   - with `verification.logicTests: \"required\"`, stop and point to `\u002Ftests` if\n     logic changed but no test runner is configured\n3. If UI is involved, inspect the running app when possible. Prefer Playwright if\n   it is already installed or declared. Capture screenshots when they add useful\n   evidence. Check for console errors and failed requests.\n   With `verification.uiEvidence: \"required\"`, direct browser evidence is\n   mandatory and unavailable evidence is a hard stop.\n4. Self-review the diff for the step:\n   - does it match the spec?\n   - did it add scope?\n   - is the error path handled?\n   - did it follow `coding-standards.md`?\n   - are tests present for new in-scope logic when the test gate is on?\n5. Fix obvious issues and rerun the failed checks.\n6. Mark the step checked in `current-feature.md` only after the step passes.\n7. When `workflow.checkpointCommits: \"enabled\"`, create a checkpoint commit on\n   the feature or fix branch for the passing step. Include the code, tests, and\n   the updated `current-feature.md` checkbox. Use a conventional message such as\n   `feat: checkpoint mock snapshot route` or `fix: checkpoint stale service\n   filter`. Keep the message about the step, not about Autopilot. When the value\n   is `disabled`, leave the passing step uncommitted for the final review.\n\nDo not batch the whole feature into one large diff. If a step gets too large,\nsplit the step in `current-feature.md` and continue with the first smaller step.\n\n## Step 5 - configured acceptance check\n\nAfter all implementation steps are checked, apply\n`qualityGates.regular.check`:\n\n- `manual` - skip the automatic `\u002Fcheck`; it remains available when explicitly\n  requested.\n- `when-behavioral` - run `\u002Fcheck` when a done-when needs observed runtime\n  behavior such as a click, request, CLI command, download, background job, or\n  multi-screen flow.\n- `always` - run `\u002Fcheck` for every work item.\n\nAfter the required verification and configured Check gate pass, set the current\nspec's `**Status:**` to `verified` before any independent-review checkpoint.\nAudit findings and review state may still block completion. Leave the spec `in\nprogress`, `verification failed`, or `verification incomplete` on any stop that\nlacks complete verification evidence.\n\nFor pure library or CLI work, build plus tests and representative command output\nmay be enough. Be explicit about the evidence used.\n\n## Step 6 - configured quality audit and repair\n\nApply `qualityGates.regular.independentReview` before a same-session audit:\n\n- `manual` skips automatic independent review unless a request already exists.\n- `when-sensitive` requires it for authentication, authorization, payments,\n  secrets, personal or user data, migrations, destructive operations, external\n  side effects, security boundaries, or unusually broad changes.\n- `always` requires it for every work item.\n\nThe selected review runs after all implementation steps, final Verify, required\nCheck, and the verified spec, before the final review packet and `\u002Fcomplete`.\n`review.independentExecution` chooses the manual fresh-session handoff or an\nautomatic isolated reviewer; it does not change when the gate is selected.\n\nWhen selected, do not review the builder's work in this session. Ensure\napplication code is in an approved clean checkpoint. Include the verified spec\nwhen tracked; an intentionally ignored spec uses Audit's local `Spec snapshot`\ncontract without changing visibility. Then follow Phase A of\n`\u002Faudit independent current`. With automatic execution, spawn and wait for the\nisolated reviewer, then validate its normal receipt. With manual execution, stop\nwith the handoff. Autopilot may use its existing configured checkpoint authority\nwhen checkpoint commits are enabled; otherwise show the exact review-checkpoint\ncandidate and ask before committing. On resume, continue only when a fresh\nreviewer wrote a current `passed` receipt. Repair `changes-requested` P0\u002FP1\nfindings within the normal scope and attempt limit, then obtain a new checkpoint\nand prepare a new review. A passing independent receipt satisfies the configured\nAudit gate. A local-spec-only revision may reuse the same approved product HEAD\nafter normal spec and verification gates, with a new snapshot\u002Frequest and full\nfresh review. Do not create an empty commit for ignored spec changes. Automatic\nexecution must also confirm access to the same local spec\u002Fsnapshot and installed\nskills; otherwise retain the request and use the manual handoff in the original\ncheckout.\n\nThe request records `Requested execution`; the receipt records `Actual\nexecution`. Require the execution and reviewer-context pairing defined by the\nproject-local review contract, including actual manual plus `fresh session` when\nan automatic request explicitly falls back.\nOn resume, a pending request without `Requested execution` is legacy manual-only.\nNever add execution fields or run a subagent against it.\n\nApply `qualityGates.regular.audit`:\n\n- `manual` - skip the automatic audit; `\u002Faudit` remains available when\n  explicitly requested.\n- `when-sensitive` - run `\u002Faudit current` for authentication, authorization,\n  payments, secrets, personal or user data, migrations, destructive operations,\n  external side effects, security boundaries, or unusually broad changes.\n- `always` - run `\u002Faudit current` for every work item.\n\nWhen the gate runs, audit the active feature, its diff, and the nearby code\naffected by the change. This is a targeted feature audit, not a repository-wide\ncleanup pass. Findings are recorded in `blueprint\u002Fcontext\u002Ffindings.md` with\ndurable IDs and statuses, as `\u002Faudit` defines; the ledger reports status and\nnever scopes what the audit examines.\n\nFor every finding:\n\n1. Validate it against the actual code, spec, tests, `coding-standards.md`, and\n   local project patterns. An audit finding is evidence to investigate, not an\n   automatic instruction to edit.\n2. Repair confirmed P0 and P1 findings when the fix stays inside the approved\n   feature scope, does not require a product or architecture decision, and does\n   not remove or change shipped behavior. Set the repaired finding to `fixed` in\n   the ledger, never `closed`.\n3. Report P2 and P3 findings in the final packet. Fix them only when the change\n   is small, directly caused by the current feature, and clearly required by the\n   project standards.\n4. If a confirmed P0 or P1 finding cannot be repaired safely within scope, stop\n   and report it. Do not present the feature as ready for `\u002Fcomplete`.\n\nAfter any audit repair:\n\n1. Rerun the documented `Verify` command when present; otherwise rerun the\n   affected build, lint, typecheck, and test commands.\n2. Rerun the acceptance evidence affected by the repair.\n3. Recheck the repaired area using the same targeted audit criteria. When that\n   recheck confirms the original defect is gone and the repair introduced no\n   new one, move the `fixed` finding to `closed` under the `\u002Faudit` close\n   conditions and name it in the packet. An unrelated new finding gets its own\n   ledger entry and does not keep the repaired one open.\n4. Create a checkpoint commit only after the repair and its checks pass, and\n   only when checkpoint commits are enabled.\n\nUse the existing two-attempt hard stop for repeated repair failures. Do not widen\nthe feature into a general refactor, silently suppress a finding, or turn this\nstep into a full-project hardening pass. A broader cleanup remains a separate\n`\u002Faudit` followed by planned `\u002Ffix` work.\n\n## Step 7 - configured try guide\n\nApply `qualityGates.regular.tryGuide`:\n\n- `manual` - skip automatic generation; `\u002Fcheck guide` remains available when\n  explicitly requested.\n- `when-user-facing` - run `\u002Fcheck guide` when the change affects UI,\n  navigation, copy, a public API or CLI, output, or another workflow a person\n  directly uses.\n- `always` - generate a guide for every work item.\n\nThe guide is a review artifact, not proof. Never claim the user performed it.\n\n## Step 8 - final review packet\n\nStop with a concise review packet. Keep it useful enough for `\u002Fcomplete` but not\na full audit report:\n\n- branch name\n- target feature or fix\n- whether the spec was created or resumed\n- what the spec critique changed\n- changed files and why each changed\n- build\u002Ftest\u002Fcheck commands run, with pass or fail\n- effective regular quality-gate policies and which automatic gates ran or were\n  skipped\n- independent-review target, selected reviewer and model, and receipt state\n- screenshots or output paths, when relevant\n- how to try it manually, or a pointer to `\u002Fcheck guide` for the full walkthrough\n- checkpoint commits created\n- self-review findings\n- targeted audit scope and findings, when the audit gate ran\n- audit repairs made and checks rerun, when applicable\n- P0\u002FP1 findings still `open` or `fixed` in `blueprint\u002Fcontext\u002Ffindings.md`,\n  which block `\u002Fcomplete`\n- unresolved risks or skipped checks\n- exact next action\n\nIf everything is green, the next action is usually: review the diff, run `\u002Fcheck guide`\nif its gate was manual and a walkthrough is wanted, then `\u002Fcomplete`.\n\nAlways offer a read-only walkthrough of the completed code after the packet.\nFollow the spec's build steps, explain the key files, symbols, flow, and\nnon-obvious decisions, then offer a focused deep dive. Keep `\u002Fcheck guide` distinct as\nthe manual product-review path.\n\nIf something failed, name the failing check and the next fix target.\n\n## Hard Stops\n\nStop immediately and report instead of continuing when Autopilot would need to:\n\n- commit on `main`, merge, delete a branch, push, deploy, publish, or send\n  anything\n- delete data, reset a database, run irreversible migrations, kill processes, or\n  change system settings\n- install dependencies or use network access without the current tool's approval\n  flow\n- make a product decision not covered by the docs\n- continue after two failed fix attempts on the same issue\n- hide, skip, or hand-wave a failing check\n\n## Rules\n\n- One Autopilot run handles one feature or one fix.\n- Autopilot creates checkpoint commits on the feature or fix branch after\n  passing steps only when project config enables them.\n- Autopilot uses `qualityGates.regular`. The Continuous gate policies and the\n  `continuous` section do not change an Autopilot run.\n- When its audit gate runs, Autopilot audits the active feature and affected\n  code, not the entire project.\n- A P0 or P1 finding left `open` or `fixed` in `blueprint\u002Fcontext\u002Ffindings.md`\n  blocks readiness for `\u002Fcomplete`. The ledger is what makes this enforceable.\n- Autopilot stops before `\u002Fcomplete`. It never merges.\n- The Blueprint files remain the state machine. Keep\n  `current-feature.md` accurate as steps complete.\n- Follow `coding-standards.md`, `ai-interaction.md`, and `AGENTS.md`.\n- Prefer fewer, higher-quality changes over broad coverage.\n- Report uncertainty plainly. A blocked run is useful if it tells the truth.\n\n## Formatting\n\nFormat the output to match the project's conventions in\n`blueprint\u002Fcontext\u002Fai-interaction.md`: concise, scannable markdown, with lists for\nenumerations and tables for matrices rather than dense paragraphs.\n","# autopilot - optional Blueprint loop\n\n**Context reuse:** Reuse any required file already loaded in project instructions or the current session. Read it again only if absent, changed, or exact current bytes or line references are needed.\n\n**First action:** Before project inspection, preflight, or any other tool call,\npublish `running` to `blueprint\u002F.state\u002Frun.json` using the dashboard activity\ncontract in `AGENTS.md`.\n\nWhere this sits in the workflow:\n\n    \u002Fstatus  ->  [autopilot]  ->  review packet  ->  \u002Fcomplete\n    (where       (spec, build,      (human review,    (log, commit,\n     are we?)     configured gates)  fixes if needed)  merge with approval)\n\nAutopilot is an explicit opt-in path. It uses the same Blueprint files and the\nsame quality gates, but it does not stop after every normal review point. A\nsingle user request is permission to run one bounded loop until the feature is\nready for review, blocked, or unsafe to continue.\n\nIt combines `\u002Ffeature` or `\u002Ffix` with `\u002Fimplement` and continues through the\nspec-review stop retained by the normal workflow. That human spec approval is\nthe main control Autopilot intentionally removes. It does not remove the final\nreview packet or the option to walk through the completed code.\n\nIt does **not** replace the normal workflow. `\u002Ffeature`, `\u002Fimplement`, `\u002Fcheck`,\nand `\u002Fcomplete` remain the conservative default.\n\nDo not suggest Autopilot as the default next action. Use it only when the user\nexplicitly asks for it.\n\nThe explicit Autopilot request is permission to create checkpoint commits on the\nfeature or fix branch after passing implementation steps when\n`workflow.checkpointCommits` is enabled. It is not permission to merge, push,\ndeploy, publish, send, delete data, or run destructive actions.\n\n## Input\n\nCommon forms:\n\n- No argument: resume the current feature if one exists, otherwise target the next\n  unchecked build-plan item.\n- A number or name: target that build-plan feature, for example `\u002Fautopilot 3` or\n  `$autopilot \"directory listing\"`.\n- `fix \"\u003Cissue>\"`: write and build an ad-hoc fix spec.\n- `resume`: continue the current feature on its existing branch.\n\nIf the requested target conflicts with a feature already in progress, stop and\nask which one should win. Do not overwrite `blueprint\u002Fcontext\u002Fcurrent-feature.md`\nsilently.\n\nRollback is intentionally excluded from Autopilot. If the request is a rollback\nor `current-feature.md` is marked `Type: Rollback`, stop and direct the user to\nthe reviewed `\u002Fimplement` path. Reversing completed work requires the explicit\ndependency and conflict gates in `\u002Frollback` and `\u002Fimplement`.\n\n## Step 1 - preflight like \u002Fstatus\n\nRead the same state `\u002Fstatus` reads:\n\n- `AGENTS.md`\n- `blueprint\u002Fconfig.json`\n- `blueprint\u002Fproject-plan.md`\n- `blueprint\u002Fbuild-plan.md`\n- `blueprint\u002Fcontext\u002Fproject-overview.md`\n- `blueprint\u002Fcontext\u002Fcurrent-feature.md`\n- `blueprint\u002Fcontext\u002Ffindings.md`\n- `blueprint\u002Fcontext\u002Fcoding-standards.md`\n- `blueprint\u002Fcontext\u002Fai-interaction.md`\n- git branch, status, and recent log\n\nThen decide whether it is safe to run.\n\nStop before changing files when:\n\n- `blueprint\u002Fconfig.json` exists but is invalid. Point the user to `\u002Fdoctor`.\n- The repo is not a git repo.\n- The working tree is dirty and there is no current feature tying those changes\n  to this run.\n- `current-feature.md` has real work and the user requested a different target.\n- `project-overview.md` is missing or stale and the planning docs are not clear\n  enough to regenerate it.\n- The next feature is visual or replication-heavy and no design reference exists.\n- The task needs product, data, auth, billing, or destructive decisions the docs\n  do not answer.\n\nIf the only issue is that `project-overview.md` is stale and the plans are clear,\nregenerate it using the `\u002Foverview` behavior and continue. Include that in the\nfinal packet.\n\nThe initial `blueprint\u002F.state\u002Frun.json` record required by `AGENTS.md` must\nalready show command `autopilot` and status `running` before preflight begins.\nAfter preflight passes, enrich it with boundary `reviewed`, the target feature\nor fix, and build-step progress when known. Update it after the spec, each\npassing build step, and each configured gate. On a hard\nstop, set status `blocked` with `\u002Fautopilot resume` when resuming is safe. At the\nfinal review packet, set status `ready` because Autopilot stops before\n`\u002Fcomplete`. Activity reporting must never weaken or block the workflow itself.\n\n## Step 2 - choose or write the spec\n\nIf `blueprint\u002Fcontext\u002Fcurrent-feature.md` already contains an active spec,\nresume it. Read checked steps and continue from the first unchecked step.\n\nIf there is no active spec:\n\n1. Use the `\u002Ffeature` behavior for a planned feature, or `\u002Ffix` behavior for a\n   requested fix.\n2. Write `blueprint\u002Fcontext\u002Fcurrent-feature.md`.\n3. Red-team the spec before building:\n   - missing unhappy paths\n   - oversized steps\n   - undefined contracts\n   - missing design reference\n   - scope creep\n   - vague done-whens\n   - missing testing plan when `AGENTS.md` declares a test command\n4. Apply the spec fixes.\n\nAutopilot may continue past this spec gate because the user explicitly invoked\nAutopilot. Still report what the critique changed in the final packet.\nFollow the proportional-engineering contract in `AGENTS.md` throughout this run.\n\n## Step 3 - create or reuse the branch\n\nUse the same branch rules as `\u002Fimplement`:\n\n- Feature: the configured feature prefix, default `feature\u002F\u003Cname>`\n- Fix: the configured fix prefix, default `fix\u002F\u003Cname>`\n\nIf the branch already exists, switch to it only if it matches the active spec.\nIf switching branches would strand unrelated dirty work, stop and report the\nproblem.\n\n## Step 4 - implement in small steps\n\nWork through the spec's build steps in order. Each step must remain reviewable.\nDo not pause for user approval after each passing step, regardless of the\nconfigured `workflow.stepReview` value. The review happens at the final packet\nunless a hard stop is hit.\n\nFor every step:\n\n1. Implement only that step.\n2. Run the relevant verification:\n   - the exact `Verify` command from `AGENTS.md`, when declared\n   - otherwise the build, relevant test, lint, and typecheck commands already\n     documented by the project\n   - browser, CLI, API, or app-level evidence for behavioral done-whens\n   - with `verification.logicTests: \"required\"`, stop and point to `\u002Ftests` if\n     logic changed but no test runner is configured\n3. If UI is involved, inspect the running app when possible. Prefer Playwright if\n   it is already installed or declared. Capture screenshots when they add useful\n   evidence. Check for console errors and failed requests.\n   With `verification.uiEvidence: \"required\"`, direct browser evidence is\n   mandatory and unavailable evidence is a hard stop.\n4. Self-review the diff for the step:\n   - does it match the spec?\n   - did it add scope?\n   - is the error path handled?\n   - did it follow `coding-standards.md`?\n   - are tests present for new in-scope logic when the test gate is on?\n5. Fix obvious issues and rerun the failed checks.\n6. Mark the step checked in `current-feature.md` only after the step passes.\n7. When `workflow.checkpointCommits: \"enabled\"`, create a checkpoint commit on\n   the feature or fix branch for the passing step. Include the code, tests, and\n   the updated `current-feature.md` checkbox. Use a conventional message such as\n   `feat: checkpoint mock snapshot route` or `fix: checkpoint stale service\n   filter`. Keep the message about the step, not about Autopilot. When the value\n   is `disabled`, leave the passing step uncommitted for the final review.\n\nDo not batch the whole feature into one large diff. If a step gets too large,\nsplit the step in `current-feature.md` and continue with the first smaller step.\n\n## Step 5 - configured acceptance check\n\nAfter all implementation steps are checked, apply\n`qualityGates.regular.check`:\n\n- `manual` - skip the automatic `\u002Fcheck`; it remains available when explicitly\n  requested.\n- `when-behavioral` - run `\u002Fcheck` when a done-when needs observed runtime\n  behavior such as a click, request, CLI command, download, background job, or\n  multi-screen flow.\n- `always` - run `\u002Fcheck` for every work item.\n\nAfter the required verification and configured Check gate pass, set the current\nspec's `**Status:**` to `verified` before any independent-review checkpoint.\nAudit findings and review state may still block completion. Leave the spec `in\nprogress`, `verification failed`, or `verification incomplete` on any stop that\nlacks complete verification evidence.\n\nFor pure library or CLI work, build plus tests and representative command output\nmay be enough. Be explicit about the evidence used.\n\n## Step 6 - configured quality audit and repair\n\nApply `qualityGates.regular.independentReview` before a same-session audit:\n\n- `manual` skips automatic independent review unless a request already exists.\n- `when-sensitive` requires it for authentication, authorization, payments,\n  secrets, personal or user data, migrations, destructive operations, external\n  side effects, security boundaries, or unusually broad changes.\n- `always` requires it for every work item.\n\nThe selected review runs after all implementation steps, final Verify, required\nCheck, and the verified spec, before the final review packet and `\u002Fcomplete`.\n`review.independentExecution` chooses the manual fresh-session handoff or an\nautomatic isolated reviewer; it does not change when the gate is selected.\n\nWhen selected, do not review the builder's work in this session. Ensure\napplication code is in an approved clean checkpoint. Include the verified spec\nwhen tracked; an intentionally ignored spec uses Audit's local `Spec snapshot`\ncontract without changing visibility. Then follow Phase A of\n`\u002Faudit independent current`. With automatic execution, spawn and wait for the\nisolated reviewer, then validate its normal receipt. With manual execution, stop\nwith the handoff. Autopilot may use its existing configured checkpoint authority\nwhen checkpoint commits are enabled; otherwise show the exact review-checkpoint\ncandidate and ask before committing. On resume, continue only when a fresh\nreviewer wrote a current `passed` receipt. Repair `changes-requested` P0\u002FP1\nfindings within the normal scope and attempt limit, then obtain a new checkpoint\nand prepare a new review. A passing independent receipt satisfies the configured\nAudit gate. A local-spec-only revision may reuse the same approved product HEAD\nafter normal spec and verification gates, with a new snapshot\u002Frequest and full\nfresh review. Do not create an empty commit for ignored spec changes. Automatic\nexecution must also confirm access to the same local spec\u002Fsnapshot and installed\nskills; otherwise retain the request and use the manual handoff in the original\ncheckout.\n\nThe request records `Requested execution`; the receipt records `Actual\nexecution`. Require the execution and reviewer-context pairing defined by the\nproject-local review contract, including actual manual plus `fresh session` when\nan automatic request explicitly falls back.\nOn resume, a pending request without `Requested execution` is legacy manual-only.\nNever add execution fields or run a subagent against it.\n\nApply `qualityGates.regular.audit`:\n\n- `manual` - skip the automatic audit; `\u002Faudit` remains available when\n  explicitly requested.\n- `when-sensitive` - run `\u002Faudit current` for authentication, authorization,\n  payments, secrets, personal or user data, migrations, destructive operations,\n  external side effects, security boundaries, or unusually broad changes.\n- `always` - run `\u002Faudit current` for every work item.\n\nWhen the gate runs, audit the active feature, its diff, and the nearby code\naffected by the change. This is a targeted feature audit, not a repository-wide\ncleanup pass. Findings are recorded in `blueprint\u002Fcontext\u002Ffindings.md` with\ndurable IDs and statuses, as `\u002Faudit` defines; the ledger reports status and\nnever scopes what the audit examines.\n\nFor every finding:\n\n1. Validate it against the actual code, spec, tests, `coding-standards.md`, and\n   local project patterns. An audit finding is evidence to investigate, not an\n   automatic instruction to edit.\n2. Repair confirmed P0 and P1 findings when the fix stays inside the approved\n   feature scope, does not require a product or architecture decision, and does\n   not remove or change shipped behavior. Set the repaired finding to `fixed` in\n   the ledger, never `closed`.\n3. Report P2 and P3 findings in the final packet. Fix them only when the change\n   is small, directly caused by the current feature, and clearly required by the\n   project standards.\n4. If a confirmed P0 or P1 finding cannot be repaired safely within scope, stop\n   and report it. Do not present the feature as ready for `\u002Fcomplete`.\n\nAfter any audit repair:\n\n1. Rerun the documented `Verify` command when present; otherwise rerun the\n   affected build, lint, typecheck, and test commands.\n2. Rerun the acceptance evidence affected by the repair.\n3. Recheck the repaired area using the same targeted audit criteria. When that\n   recheck confirms the original defect is gone and the repair introduced no\n   new one, move the `fixed` finding to `closed` under the `\u002Faudit` close\n   conditions and name it in the packet. An unrelated new finding gets its own\n   ledger entry and does not keep the repaired one open.\n4. Create a checkpoint commit only after the repair and its checks pass, and\n   only when checkpoint commits are enabled.\n\nUse the existing two-attempt hard stop for repeated repair failures. Do not widen\nthe feature into a general refactor, silently suppress a finding, or turn this\nstep into a full-project hardening pass. A broader cleanup remains a separate\n`\u002Faudit` followed by planned `\u002Ffix` work.\n\n## Step 7 - configured try guide\n\nApply `qualityGates.regular.tryGuide`:\n\n- `manual` - skip automatic generation; `\u002Fcheck guide` remains available when\n  explicitly requested.\n- `when-user-facing` - run `\u002Fcheck guide` when the change affects UI,\n  navigation, copy, a public API or CLI, output, or another workflow a person\n  directly uses.\n- `always` - generate a guide for every work item.\n\nThe guide is a review artifact, not proof. Never claim the user performed it.\n\n## Step 8 - final review packet\n\nStop with a concise review packet. Keep it useful enough for `\u002Fcomplete` but not\na full audit report:\n\n- branch name\n- target feature or fix\n- whether the spec was created or resumed\n- what the spec critique changed\n- changed files and why each changed\n- build\u002Ftest\u002Fcheck commands run, with pass or fail\n- effective regular quality-gate policies and which automatic gates ran or were\n  skipped\n- independent-review target, selected reviewer and model, and receipt state\n- screenshots or output paths, when relevant\n- how to try it manually, or a pointer to `\u002Fcheck guide` for the full walkthrough\n- checkpoint commits created\n- self-review findings\n- targeted audit scope and findings, when the audit gate ran\n- audit repairs made and checks rerun, when applicable\n- P0\u002FP1 findings still `open` or `fixed` in `blueprint\u002Fcontext\u002Ffindings.md`,\n  which block `\u002Fcomplete`\n- unresolved risks or skipped checks\n- exact next action\n\nIf everything is green, the next action is usually: review the diff, run `\u002Fcheck guide`\nif its gate was manual and a walkthrough is wanted, then `\u002Fcomplete`.\n\nAlways offer a read-only walkthrough of the completed code after the packet.\nFollow the spec's build steps, explain the key files, symbols, flow, and\nnon-obvious decisions, then offer a focused deep dive. Keep `\u002Fcheck guide` distinct as\nthe manual product-review path.\n\nIf something failed, name the failing check and the next fix target.\n\n## Hard Stops\n\nStop immediately and report instead of continuing when Autopilot would need to:\n\n- commit on `main`, merge, delete a branch, push, deploy, publish, or send\n  anything\n- delete data, reset a database, run irreversible migrations, kill processes, or\n  change system settings\n- install dependencies or use network access without the current tool's approval\n  flow\n- make a product decision not covered by the docs\n- continue after two failed fix attempts on the same issue\n- hide, skip, or hand-wave a failing check\n\n## Rules\n\n- One Autopilot run handles one feature or one fix.\n- Autopilot creates checkpoint commits on the feature or fix branch after\n  passing steps only when project config enables them.\n- Autopilot uses `qualityGates.regular`. The Continuous gate policies and the\n  `continuous` section do not change an Autopilot run.\n- When its audit gate runs, Autopilot audits the active feature and affected\n  code, not the entire project.\n- A P0 or P1 finding left `open` or `fixed` in `blueprint\u002Fcontext\u002Ffindings.md`\n  blocks readiness for `\u002Fcomplete`. The ledger is what makes this enforceable.\n- Autopilot stops before `\u002Fcomplete`. It never merges.\n- The Blueprint files remain the state machine. Keep\n  `current-feature.md` accurate as steps complete.\n- Follow `coding-standards.md`, `ai-interaction.md`, and `AGENTS.md`.\n- Prefer fewer, higher-quality changes over broad coverage.\n- Report uncertainty plainly. A blocked run is useful if it tells the truth.\n\n## Formatting\n\nFormat the output to match the project's conventions in\n`blueprint\u002Fcontext\u002Fai-interaction.md`: concise, scannable markdown, with lists for\nenumerations and tables for matrices rather than dense paragraphs.\n",{"name":5,"description":7,"disable-model-invocation":24},true,"main","0007775402258962018a93be4dbf1cbcbc1536d8",".claude\u002Fskills\u002Fautopilot",3,"https:\u002F\u002Fgithub.com\u002Fbradtraversy\u002Fskillpass","https:\u002F\u002Fgithub.com\u002Fbradtraversy","traversymedia.com","Full stack web developer and online instructor at https:\u002F\u002Ftraversymedia.com and https:\u002F\u002Fstart.dev","2026-10-06 00:26:26",{"source":8,"skillName":5,"commands":35,"prompt":45,"zipAvailable":24},[36,39,42],{"tool":37,"command":38},"npx","npx skills add bradtraversy\u002Fskillpass --skill autopilot",{"tool":40,"command":41},"bunx","bunx skills add bradtraversy\u002Fskillpass --skill autopilot",{"tool":43,"command":44},"pnpm","pnpm dlx skills add bradtraversy\u002Fskillpass --skill autopilot","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fbradtraversy\u002Fskillpass\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fautopilot\u002FSKILL.md",[],[48,54,60,65,71,77],{"skillNo":49,"name":50,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":51,"categories":52,"keywords":53,"publishedAt":19},"01M782QBBKH3WTBZDE9DP52XH4","plan","https:\u002F\u002Fgithub.com\u002Fbradtraversy\u002Fskillpass\u002Fblob\u002Fmain\u002Fpackages\u002Fvalidator\u002Ffixtures\u002Fworkflow-pack\u002Fskills\u002Fplan\u002FSKILL.md",[],[],{"skillNo":55,"name":56,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":57,"categories":58,"keywords":59,"publishedAt":19},"01M782Q980Y1KP1J77TNJHD5H8","apply","https:\u002F\u002Fgithub.com\u002Fbradtraversy\u002Fskillpass\u002Fblob\u002Fmain\u002Fpackages\u002Fvalidator\u002Ffixtures\u002Fworkflow-pack\u002Fskills\u002Fapply\u002FSKILL.md",[],[],{"skillNo":61,"name":50,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":62,"categories":63,"keywords":64,"publishedAt":19},"01M782Q75AS3P8SW7D819TACVD","https:\u002F\u002Fgithub.com\u002Fbradtraversy\u002Fskillpass\u002Fblob\u002Fmain\u002Fpackages\u002Fvalidator\u002Ffixtures\u002Fworkflow-pack\u002F.claude\u002Fskills\u002Fplan\u002FSKILL.md",[],[],{"skillNo":66,"name":67,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":68,"categories":69,"keywords":70,"publishedAt":19},"01M782Q5WAP9DQH14TEKG5ZFX5","undeclared-network","https:\u002F\u002Fgithub.com\u002Fbradtraversy\u002Fskillpass\u002Fblob\u002Fmain\u002Fpackages\u002Fvalidator\u002Ffixtures\u002Fundeclared-network\u002FSKILL.md",[],[],{"skillNo":72,"name":73,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":74,"categories":75,"keywords":76,"publishedAt":19},"01M782Q424YF4TAYYX9TCCZCS8","prompt-injection","https:\u002F\u002Fgithub.com\u002Fbradtraversy\u002Fskillpass\u002Fblob\u002Fmain\u002Fpackages\u002Fvalidator\u002Ffixtures\u002Fprompt-injection\u002FSKILL.md",[],[],{"skillNo":78,"name":79,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":80,"categories":81,"keywords":82,"publishedAt":19},"01M782Q28RQV6MJM98J6EN7XFF","missing-manifest","https:\u002F\u002Fgithub.com\u002Fbradtraversy\u002Fskillpass\u002Fblob\u002Fmain\u002Fpackages\u002Fvalidator\u002Ffixtures\u002Fmissing-manifest\u002FSKILL.md",[],[]]