[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M75FKVVCJ12997DD633S7AAQ":3,"skill-same:01M75FKVVCJ12997DD633S7AAQ":57},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":17,"publishedAt":18,"linkOnly":19,"markdownContent":20,"markdownBody":21,"frontMatter":22,"refName":23,"commitSha":24,"skillDirectory":25,"forksCount":13,"repositoryHtmlUrl":26,"repositoryHomepageUrl":6,"ownerHtmlUrl":27,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":28,"install":29,"resources":42},"01M75FKVVCJ12997DD633S7AAQ","adversarial-review",null,"Use when asked for an adversarial review, dual review, mutual-critique review, cross-critique review, or 相互批判レビュー of code (a working diff, a PR, or a whole codebase), or when a review needs higher confidence than a single reviewer gives — before a risky merge or release, or when a prior single review produced inflated severities or generic checklist findings. Also use to verify fixes made in response to such a review.","aharada54914\u002Fsdd-forge","aharada54914","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F40755043?v=4","01M75FFC4V7WXZF0PK9RYWHNGF","01M75FFC5YJJ11B7MDCRVWB9Z6",0,"MIT","https:\u002F\u002Fgithub.com\u002Faharada54914\u002Fsdd-forge\u002Fblob\u002Fmain\u002Fskills\u002Fadversarial-review\u002FSKILL.md",[],[],"2026-11-12 13:55:43",false,"---\nname: adversarial-review\ndescription: Use when asked for an adversarial review, dual review, mutual-critique review, cross-critique review, or 相互批判レビュー of code (a working diff, a PR, or a whole codebase), or when a review needs higher confidence than a single reviewer gives — before a risky merge or release, or when a prior single review produced inflated severities or generic checklist findings. Also use to verify fixes made in response to such a review.\n---\n\n# Adversarial Review (two-reviewer mutual critique)\n\n## Overview\n\nTwo reviewers with disjoint lenses review the target blind. Each then attacks\nthe other's findings with per-finding verdicts. The orchestrator synthesizes\none report, and fixes are later verified by a fresh-context reviewer. The\ncross-critique and the fresh verification are what catch severity inflation,\ngeneric-checklist findings, over-engineered fixes, and errors inside the fixes\nthemselves — failure modes a single review cannot see.\n\n**Core principle: every claim needs file:line evidence, and silence must be\nauditable — reviewers state what they checked and found clean.**\n\n## When to Use\n\n- Pre-merge review of a risky diff or PR; audit of a codebase or subsystem\n- After such a review, to verify the fixes (Phase R)\n- NOT for small routine diffs — a single review is cheaper and enough\n- NOT for sdd-forge SDD gate reviews (spec\u002Fimpl\u002Ftask-review-loop have their\n  own deterministic contracts)\n\n## Protocol\n\nCopy-paste reviewer prompts: [references\u002Freviewer-prompts.md](references\u002Freviewer-prompts.md).\nReport structure: [templates\u002Freport-template.md](templates\u002Freport-template.md).\n\n| Phase | Actor | What happens |\n|-------|-------|--------------|\n| 0 Scope | Orchestrator | Pin the target (diff \u002F PR \u002F paths) and commit hash; gather codebase context (scale, deployment reality, house rules such as AGENTS.md); note optional focus areas; fill the prompt placeholders. |\n| 1 Blind review | Reviewers A + B in parallel | A = design \u002F maintainability \u002F readability (senior architect). B = security \u002F testing \u002F operational risk. Spawn both with the Agent tool in ONE message (background). Neither sees the other's output. |\n| 2 Cross-critique | Same two agents, resumed | SendMessage each reviewer the other's full report verbatim (context preserved — no re-reading cost). Per finding: SUPPORT \u002F PROPOSE-SEVERITY-CHANGE \u002F PROPOSE-REJECT \u002F SUPPLEMENT, each with file:line evidence. Fix proposals are critiqued separately from findings. New findings get IDs A-C1…, B-C1… |\n| 3 Synthesis | Orchestrator (main context) | Adopt or reject every finding with a stated reason; set final severity; merge converging IDs (e.g. `A-6+B-C2`); union the verified non-findings; write the phased remediation plan with a TODO checklist. Rejected findings stay in the report with the rejection reason. |\n| R Fix verification | NEW agent, fresh context | After fixes land: spawn a reviewer with no prior involvement. Input = adopted findings + the fix diff only. It re-verifies each fix at file:line and checks every factual claim in the fix text (counts, ratios, \"all N sites\"). Never verify fixes in the context that wrote them. |\n\n## Iron rules — in every reviewer prompt, never trimmed\n\nThe prompt templates already contain these; they are the source of the\nprotocol's effectiveness.\n\n1. **file:line evidence only** — cite only lines actually read this session; a\n   finding without evidence is invalid, not \"low confidence\"\n2. **Verified non-findings are mandatory** — every report ends with the\n   checked-and-clean list (e.g. \"no hardcoded secrets; CI permissions sound\")\n3. **Context-grounded severity** — CRITICAL\u002FHIGH require a concrete harm path\n   in THIS codebase's real context; importing generic-checklist requirements\n   (web-scale demands on an internal desktop tool) is itself a critique target\n4. **Proportionate fixes** — an over-engineered fix (a DI container for one\n   seam) is a critique target even when the finding stands\n5. **No subagents** — reviewers do all reading and analysis themselves, in\n   their own context; they never delegate to background agents\n6. **Read-only** — reviewers modify nothing and return the report as their\n   final message\n\n## Recovery procedures\n\n| Symptom | Action |\n|---------|--------|\n| Reviewer delegated to its own background agent and stalled | SendMessage the reviewer: \"Do not delegate. Do the review yourself in this context, now.\" |\n| Reviewer died mid-run (e.g. API Overloaded) | SendMessage-resume the same agent — the transcript survives and work is not lost. Do not respawn fresh; that loses its Phase 1 context. |\n| Malformed or incomplete report | One SendMessage re-ask against the output contract |\n\n## Common mistakes\n\n- Verifying fixes in the authoring context — in the proving run, a\n  ratio-inversion error inside a fix text (\"2\u002F5\" vs \"3\u002F5\" lock sites) was\n  caught only by the fresh round-2 reviewer\n- Dropping the design\u002Fmaintainability lens — orchestrators left to themselves\n  converge on correctness-vs-security and produce no maintainability findings\n- Treating cross-critique as consensus-seeking — \"plausible but unverified\" is\n  not SUPPORT\n- Silently dropping rejected findings — keep them in the report with the\n  reason, so the same false alarm is not re-raised later\n- Skipping Phase R because \"the fixes were straightforward\"\n\n## Evaluation record (ADR-0027 risk-adaptive lane)\n\nBefore using a report as `Adversarial-Lane: fired`, verify its currentness with\nthe read-only CLI (Node and `mcp\u002Fsdd-forge-mcp` dependencies must be installed):\n\n```sh\nnode mcp\u002Fsdd-forge-mcp\u002Fscripts\u002Fcheck-adversarial-report.mjs --repo . --report \u002FABSOLUTE\u002FEVIDENCE-CHECKOUT\u002Freports\u002Fadversarial-review\u002FBRANCH\u002Freport.md --base origin\u002Fmain --report-sha256 SAVED_REPORT_SHA256\n```\n\nAt review completion, record the SHA-256 of the final report bytes separately\nin the review receipt or PR metadata as `report_sha256`. Pass that saved digest\nabove; do not recompute it from the current report to make verification pass.\nThe digest is external to avoid a self-referential hash inside the report.\nRecord `diff_sha256` from the bytes of\n`git diff --binary --no-ext-diff --no-textconv MERGE_BASE_SHA..HEAD_SHA`.\nExit 0 and `status: current` are required; nonzero (including unavailable Git,\nmissing files or invalid metadata) cannot satisfy the lane. This command does\nnot launch agents, change review verdicts, or modify repository files. The same\ncommand and JSON output apply in Bash and PowerShell.\n\n### Publish evidence without moving the reviewed HEAD\n\nKeep the final report and its companion files outside the reviewed branch.\nThe `reports\u002Fadversarial-review\u002F\u003Cbranch-slug>\u002F` path is relative to a separate\nevidence checkout, not a requirement to commit evidence on the target branch.\nUse a separate evidence repository or a separate evidence branch\u002Fworktree that\nwill not be merged into the target before its currentness check.\n\n1. Finish and commit the target changes; record that exact HEAD, merge base,\n   and full diff digest for the review. Do not modify them to fit a report.\n2. Produce the report and any `evaluation.json`\u002Fcross-critique annex under the\n   canonical path in the evidence checkout. Save the final report digest in a\n   separate receipt before publication.\n3. Commit\u002Fpublish the evidence there. In the target PR, link to the report at\n   the immutable **evidence commit**, and record the target HEAD, merge base,\n   and saved report digest. Do not use a mutable evidence-branch URL.\n4. To verify, retrieve that evidence commit, then pass its absolute report\n   path to `--report` and the reviewed checkout to `--repo`, as above. Verify\n   against the target PR's current HEAD and base; do not check out an old target\n   commit merely to obtain a passing result.\n\nCommitting a report or receipt on the target branch changes HEAD and makes the\nold report stale, even if no product code changed. There is no evidence-only\ncommit exemption. Historical reports may remain in the target repository but\ncannot claim currentness. A target update requires a fresh review of the new\ntarget, not editing the old report's hashes. Publishing evidence separately\navoids a circular report\u002Fcommit hash without excluding any target diff files.\n\nWhen invoked as the ADR-0027 pre-PR lane (i.e. the firing predicate is\nsatisfied), the orchestrator produces an `evaluation.json` alongside `report.md`\nin the evidence checkout's `reports\u002Fadversarial-review\u002F\u003Cbranch-slug>\u002F`.\nNon-triggering invocations\nproduce no `evaluation.json`.\n\n`evaluation.json` must conform to\n`contracts\u002Fadversarial-review-evaluation.v1.schema.json`. Fill all required\nfields at the end of Phase 3 (synthesis). Update `phase_r.*` after Phase R\ncompletes.\n\nEvery new evaluation must include `token_usage`. Use a measured `total_tokens`\nand its telemetry `source` only when the host reports usage for the complete\nreview run (including resumed\u002Frecovery calls). Otherwise record `total_tokens:\nnull` and a nonblank `unavailable_reason`, including partial telemetry as an\nunavailability reason. Never estimate tokens or substitute zero. This is an\nobservability record, not a gate failure or a verdict input. Older v1 records\nwithout this field remain readable; do not retrofit guessed values into them.\n\nUse full 40-character commit IDs. Compute the target digest without external\ndiff drivers so the identity is reproducible across hosts:\n\n```bash\ngit diff --binary --no-ext-diff --no-textconv \"$merge_base_sha\"..\"$head_sha\" | sha256sum\n```\n\n**Minimum required fields at synthesis time:**\n\n```json\n{\n  \"schema_version\": \"adversarial-review-evaluation.v1\",\n  \"branch_slug\": \"\u003Cbranch-slug>\",\n  \"created_at\": \"\u003CISO-8601 UTC>\",\n  \"report_path\": \"reports\u002Fadversarial-review\u002F\u003Cbranch-slug>\u002Freport.md\",\n  \"trigger_reasons\": [\"workflow_surface\"],\n  \"reviewer_launch_count\": 2,\n  \"token_usage\": { \"total_tokens\": null, \"unavailable_reason\": \"host does not expose complete run token telemetry\" },\n  \"continuation_status\": { \"reviewer_a\": \"resumed\", \"reviewer_b\": \"resumed\" },\n  \"finding_counts\": { \"phase1_total\": 0, \"cross_critique_new\": 0 },\n  \"basis_counts\": { \"code_evidence\": 0, \"spec_evidence\": 0, \"concern\": 0 },\n  \"scope_counts\": { \"in_scope\": 0, \"out_of_scope\": 0, \"unclear\": 0 },\n  \"phase_r\": { \"ran\": false, \"not_ran_reason\": \"fixes not yet landed\" }\n}\n```\n\nUpdate `phase_r` to `\"ran\": true` with counts when Phase R completes.\n\n## Contracts and schemas\n\nSet the annex's root `review_lane` to `standalone-adversarial`; severity-change\nproposals use exactly `CRITICAL|HIGH|MEDIUM|LOW`. Do not convert these to the\nSDD gate vocabulary or rewrite the original reviewer verdicts. An absent lane\nretains legacy SDD gate semantics (`Critical|Major|Minor`).\n\nFor a completed annex with `verdicts: []`, record `source_finding_count: 0`\nonly after reading both original blind-review outputs and confirming they\ncontain no findings. Missing or unavailable outputs are not zero findings;\nrecord the annex as unavailable with its reason instead. A nonempty source\nreview cannot become a completed empty annex by dropping its critiques. The\ncount is an explicit declaration, not independent verification of those\noutputs; retain the original reports for review. Nonempty historical annexes\nremain readable without this field.\n\nBefore using a cross-critique annex as evidence, run the read-only checker\nagainst the final JSON file (the same command works in Bash and PowerShell):\n\n```sh\nnode mcp\u002Fsdd-forge-mcp\u002Fscripts\u002Fcheck-cross-critique.mjs \u002FABSOLUTE\u002FEVIDENCE-CHECKOUT\u002Freports\u002Fadversarial-review\u002FBRANCH\u002Fcross-critique.json\n```\n\nExit 0 with `status: valid` proves contract conformance and ordered citation\nranges, not the truth of a claim. Every evidence citation requires a nonblank\npath and claim plus positive `line_start` and `line_end`; a single-line citation\nuses equal coordinates. The checker rejects `line_end \u003C line_start`, which\nplain Draft-07 validation alone cannot enforce. Fix invalid annexes before\nusing their proposals; do not rewrite persisted reviewer verdicts or treat a\nfailed\u002Funavailable checker as successful validation. Required Node dependencies\nare the same as for the report-currentness checker above.\n\n| Schema | Purpose |\n|--------|---------|\n| `contracts\u002Fcross-critique.v1.schema.json` | cross-critique.json annex (issues #347, #348) |\n| `contracts\u002Fadversarial-review-report.v1.schema.json` | Report metadata block (issue #349) |\n| `contracts\u002Fadversarial-review-evaluation.v1.schema.json` | evaluation.json run metrics (issue #350) |\n\n## Real-world impact\n\nProving run (torque-system-manager, 2026-07-07, PR #7): cross-critique and\nfresh verification found a target service with zero production call sites\n(requirement re-scoped), two HIGH defects in TDD-complete code (unguarded\nDisposeAsync on the close path; a swap-then-dispose ordering bug), a wrong\nlock-site count inside a fix text, and re-calibrated four of five initial\nHigh\u002FMedium severities — offsetting single-review severity inflation. External\nbenchmark (2026): adversarial panels find roughly 20% more bugs than a single\nreview.\n\nUsage history: [reports\u002Fadversarial-review\u002F](..\u002F..\u002F..\u002Freports\u002Fadversarial-review\u002F)\n","# Adversarial Review (two-reviewer mutual critique)\n\n## Overview\n\nTwo reviewers with disjoint lenses review the target blind. Each then attacks\nthe other's findings with per-finding verdicts. The orchestrator synthesizes\none report, and fixes are later verified by a fresh-context reviewer. The\ncross-critique and the fresh verification are what catch severity inflation,\ngeneric-checklist findings, over-engineered fixes, and errors inside the fixes\nthemselves — failure modes a single review cannot see.\n\n**Core principle: every claim needs file:line evidence, and silence must be\nauditable — reviewers state what they checked and found clean.**\n\n## When to Use\n\n- Pre-merge review of a risky diff or PR; audit of a codebase or subsystem\n- After such a review, to verify the fixes (Phase R)\n- NOT for small routine diffs — a single review is cheaper and enough\n- NOT for sdd-forge SDD gate reviews (spec\u002Fimpl\u002Ftask-review-loop have their\n  own deterministic contracts)\n\n## Protocol\n\nCopy-paste reviewer prompts: [references\u002Freviewer-prompts.md](references\u002Freviewer-prompts.md).\nReport structure: [templates\u002Freport-template.md](templates\u002Freport-template.md).\n\n| Phase | Actor | What happens |\n|-------|-------|--------------|\n| 0 Scope | Orchestrator | Pin the target (diff \u002F PR \u002F paths) and commit hash; gather codebase context (scale, deployment reality, house rules such as AGENTS.md); note optional focus areas; fill the prompt placeholders. |\n| 1 Blind review | Reviewers A + B in parallel | A = design \u002F maintainability \u002F readability (senior architect). B = security \u002F testing \u002F operational risk. Spawn both with the Agent tool in ONE message (background). Neither sees the other's output. |\n| 2 Cross-critique | Same two agents, resumed | SendMessage each reviewer the other's full report verbatim (context preserved — no re-reading cost). Per finding: SUPPORT \u002F PROPOSE-SEVERITY-CHANGE \u002F PROPOSE-REJECT \u002F SUPPLEMENT, each with file:line evidence. Fix proposals are critiqued separately from findings. New findings get IDs A-C1…, B-C1… |\n| 3 Synthesis | Orchestrator (main context) | Adopt or reject every finding with a stated reason; set final severity; merge converging IDs (e.g. `A-6+B-C2`); union the verified non-findings; write the phased remediation plan with a TODO checklist. Rejected findings stay in the report with the rejection reason. |\n| R Fix verification | NEW agent, fresh context | After fixes land: spawn a reviewer with no prior involvement. Input = adopted findings + the fix diff only. It re-verifies each fix at file:line and checks every factual claim in the fix text (counts, ratios, \"all N sites\"). Never verify fixes in the context that wrote them. |\n\n## Iron rules — in every reviewer prompt, never trimmed\n\nThe prompt templates already contain these; they are the source of the\nprotocol's effectiveness.\n\n1. **file:line evidence only** — cite only lines actually read this session; a\n   finding without evidence is invalid, not \"low confidence\"\n2. **Verified non-findings are mandatory** — every report ends with the\n   checked-and-clean list (e.g. \"no hardcoded secrets; CI permissions sound\")\n3. **Context-grounded severity** — CRITICAL\u002FHIGH require a concrete harm path\n   in THIS codebase's real context; importing generic-checklist requirements\n   (web-scale demands on an internal desktop tool) is itself a critique target\n4. **Proportionate fixes** — an over-engineered fix (a DI container for one\n   seam) is a critique target even when the finding stands\n5. **No subagents** — reviewers do all reading and analysis themselves, in\n   their own context; they never delegate to background agents\n6. **Read-only** — reviewers modify nothing and return the report as their\n   final message\n\n## Recovery procedures\n\n| Symptom | Action |\n|---------|--------|\n| Reviewer delegated to its own background agent and stalled | SendMessage the reviewer: \"Do not delegate. Do the review yourself in this context, now.\" |\n| Reviewer died mid-run (e.g. API Overloaded) | SendMessage-resume the same agent — the transcript survives and work is not lost. Do not respawn fresh; that loses its Phase 1 context. |\n| Malformed or incomplete report | One SendMessage re-ask against the output contract |\n\n## Common mistakes\n\n- Verifying fixes in the authoring context — in the proving run, a\n  ratio-inversion error inside a fix text (\"2\u002F5\" vs \"3\u002F5\" lock sites) was\n  caught only by the fresh round-2 reviewer\n- Dropping the design\u002Fmaintainability lens — orchestrators left to themselves\n  converge on correctness-vs-security and produce no maintainability findings\n- Treating cross-critique as consensus-seeking — \"plausible but unverified\" is\n  not SUPPORT\n- Silently dropping rejected findings — keep them in the report with the\n  reason, so the same false alarm is not re-raised later\n- Skipping Phase R because \"the fixes were straightforward\"\n\n## Evaluation record (ADR-0027 risk-adaptive lane)\n\nBefore using a report as `Adversarial-Lane: fired`, verify its currentness with\nthe read-only CLI (Node and `mcp\u002Fsdd-forge-mcp` dependencies must be installed):\n\n```sh\nnode mcp\u002Fsdd-forge-mcp\u002Fscripts\u002Fcheck-adversarial-report.mjs --repo . --report \u002FABSOLUTE\u002FEVIDENCE-CHECKOUT\u002Freports\u002Fadversarial-review\u002FBRANCH\u002Freport.md --base origin\u002Fmain --report-sha256 SAVED_REPORT_SHA256\n```\n\nAt review completion, record the SHA-256 of the final report bytes separately\nin the review receipt or PR metadata as `report_sha256`. Pass that saved digest\nabove; do not recompute it from the current report to make verification pass.\nThe digest is external to avoid a self-referential hash inside the report.\nRecord `diff_sha256` from the bytes of\n`git diff --binary --no-ext-diff --no-textconv MERGE_BASE_SHA..HEAD_SHA`.\nExit 0 and `status: current` are required; nonzero (including unavailable Git,\nmissing files or invalid metadata) cannot satisfy the lane. This command does\nnot launch agents, change review verdicts, or modify repository files. The same\ncommand and JSON output apply in Bash and PowerShell.\n\n### Publish evidence without moving the reviewed HEAD\n\nKeep the final report and its companion files outside the reviewed branch.\nThe `reports\u002Fadversarial-review\u002F\u003Cbranch-slug>\u002F` path is relative to a separate\nevidence checkout, not a requirement to commit evidence on the target branch.\nUse a separate evidence repository or a separate evidence branch\u002Fworktree that\nwill not be merged into the target before its currentness check.\n\n1. Finish and commit the target changes; record that exact HEAD, merge base,\n   and full diff digest for the review. Do not modify them to fit a report.\n2. Produce the report and any `evaluation.json`\u002Fcross-critique annex under the\n   canonical path in the evidence checkout. Save the final report digest in a\n   separate receipt before publication.\n3. Commit\u002Fpublish the evidence there. In the target PR, link to the report at\n   the immutable **evidence commit**, and record the target HEAD, merge base,\n   and saved report digest. Do not use a mutable evidence-branch URL.\n4. To verify, retrieve that evidence commit, then pass its absolute report\n   path to `--report` and the reviewed checkout to `--repo`, as above. Verify\n   against the target PR's current HEAD and base; do not check out an old target\n   commit merely to obtain a passing result.\n\nCommitting a report or receipt on the target branch changes HEAD and makes the\nold report stale, even if no product code changed. There is no evidence-only\ncommit exemption. Historical reports may remain in the target repository but\ncannot claim currentness. A target update requires a fresh review of the new\ntarget, not editing the old report's hashes. Publishing evidence separately\navoids a circular report\u002Fcommit hash without excluding any target diff files.\n\nWhen invoked as the ADR-0027 pre-PR lane (i.e. the firing predicate is\nsatisfied), the orchestrator produces an `evaluation.json` alongside `report.md`\nin the evidence checkout's `reports\u002Fadversarial-review\u002F\u003Cbranch-slug>\u002F`.\nNon-triggering invocations\nproduce no `evaluation.json`.\n\n`evaluation.json` must conform to\n`contracts\u002Fadversarial-review-evaluation.v1.schema.json`. Fill all required\nfields at the end of Phase 3 (synthesis). Update `phase_r.*` after Phase R\ncompletes.\n\nEvery new evaluation must include `token_usage`. Use a measured `total_tokens`\nand its telemetry `source` only when the host reports usage for the complete\nreview run (including resumed\u002Frecovery calls). Otherwise record `total_tokens:\nnull` and a nonblank `unavailable_reason`, including partial telemetry as an\nunavailability reason. Never estimate tokens or substitute zero. This is an\nobservability record, not a gate failure or a verdict input. Older v1 records\nwithout this field remain readable; do not retrofit guessed values into them.\n\nUse full 40-character commit IDs. Compute the target digest without external\ndiff drivers so the identity is reproducible across hosts:\n\n```bash\ngit diff --binary --no-ext-diff --no-textconv \"$merge_base_sha\"..\"$head_sha\" | sha256sum\n```\n\n**Minimum required fields at synthesis time:**\n\n```json\n{\n  \"schema_version\": \"adversarial-review-evaluation.v1\",\n  \"branch_slug\": \"\u003Cbranch-slug>\",\n  \"created_at\": \"\u003CISO-8601 UTC>\",\n  \"report_path\": \"reports\u002Fadversarial-review\u002F\u003Cbranch-slug>\u002Freport.md\",\n  \"trigger_reasons\": [\"workflow_surface\"],\n  \"reviewer_launch_count\": 2,\n  \"token_usage\": { \"total_tokens\": null, \"unavailable_reason\": \"host does not expose complete run token telemetry\" },\n  \"continuation_status\": { \"reviewer_a\": \"resumed\", \"reviewer_b\": \"resumed\" },\n  \"finding_counts\": { \"phase1_total\": 0, \"cross_critique_new\": 0 },\n  \"basis_counts\": { \"code_evidence\": 0, \"spec_evidence\": 0, \"concern\": 0 },\n  \"scope_counts\": { \"in_scope\": 0, \"out_of_scope\": 0, \"unclear\": 0 },\n  \"phase_r\": { \"ran\": false, \"not_ran_reason\": \"fixes not yet landed\" }\n}\n```\n\nUpdate `phase_r` to `\"ran\": true` with counts when Phase R completes.\n\n## Contracts and schemas\n\nSet the annex's root `review_lane` to `standalone-adversarial`; severity-change\nproposals use exactly `CRITICAL|HIGH|MEDIUM|LOW`. Do not convert these to the\nSDD gate vocabulary or rewrite the original reviewer verdicts. An absent lane\nretains legacy SDD gate semantics (`Critical|Major|Minor`).\n\nFor a completed annex with `verdicts: []`, record `source_finding_count: 0`\nonly after reading both original blind-review outputs and confirming they\ncontain no findings. Missing or unavailable outputs are not zero findings;\nrecord the annex as unavailable with its reason instead. A nonempty source\nreview cannot become a completed empty annex by dropping its critiques. The\ncount is an explicit declaration, not independent verification of those\noutputs; retain the original reports for review. Nonempty historical annexes\nremain readable without this field.\n\nBefore using a cross-critique annex as evidence, run the read-only checker\nagainst the final JSON file (the same command works in Bash and PowerShell):\n\n```sh\nnode mcp\u002Fsdd-forge-mcp\u002Fscripts\u002Fcheck-cross-critique.mjs \u002FABSOLUTE\u002FEVIDENCE-CHECKOUT\u002Freports\u002Fadversarial-review\u002FBRANCH\u002Fcross-critique.json\n```\n\nExit 0 with `status: valid` proves contract conformance and ordered citation\nranges, not the truth of a claim. Every evidence citation requires a nonblank\npath and claim plus positive `line_start` and `line_end`; a single-line citation\nuses equal coordinates. The checker rejects `line_end \u003C line_start`, which\nplain Draft-07 validation alone cannot enforce. Fix invalid annexes before\nusing their proposals; do not rewrite persisted reviewer verdicts or treat a\nfailed\u002Funavailable checker as successful validation. Required Node dependencies\nare the same as for the report-currentness checker above.\n\n| Schema | Purpose |\n|--------|---------|\n| `contracts\u002Fcross-critique.v1.schema.json` | cross-critique.json annex (issues #347, #348) |\n| `contracts\u002Fadversarial-review-report.v1.schema.json` | Report metadata block (issue #349) |\n| `contracts\u002Fadversarial-review-evaluation.v1.schema.json` | evaluation.json run metrics (issue #350) |\n\n## Real-world impact\n\nProving run (torque-system-manager, 2026-07-07, PR #7): cross-critique and\nfresh verification found a target service with zero production call sites\n(requirement re-scoped), two HIGH defects in TDD-complete code (unguarded\nDisposeAsync on the close path; a swap-then-dispose ordering bug), a wrong\nlock-site count inside a fix text, and re-calibrated four of five initial\nHigh\u002FMedium severities — offsetting single-review severity inflation. External\nbenchmark (2026): adversarial panels find roughly 20% more bugs than a single\nreview.\n\nUsage history: [reports\u002Fadversarial-review\u002F](..\u002F..\u002F..\u002Freports\u002Fadversarial-review\u002F)\n",{"name":5,"description":7},"main","0f4e3268b5403855008ae089d3c1611430a41987","skills\u002Fadversarial-review","https:\u002F\u002Fgithub.com\u002Faharada54914\u002Fsdd-forge","https:\u002F\u002Fgithub.com\u002Faharada54914","2026-11-15 02:49:38",{"source":8,"skillName":5,"commands":30,"prompt":40,"zipAvailable":41},[31,34,37],{"tool":32,"command":33},"npx","npx skills add aharada54914\u002Fsdd-forge --skill adversarial-review",{"tool":35,"command":36},"bunx","bunx skills add aharada54914\u002Fsdd-forge --skill adversarial-review",{"tool":38,"command":39},"pnpm","pnpm dlx skills add aharada54914\u002Fsdd-forge --skill adversarial-review","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Faharada54914\u002Fsdd-forge\u002Fblob\u002Fmain\u002Fskills\u002Fadversarial-review\u002FSKILL.md",true,[43,49,53],{"path":44,"resourceType":45,"resourceTypeDesc":46,"sizeBytes":47,"sourceHtmlUrl":48},"README.md","REFERENCE","参考文档",2460,"https:\u002F\u002Fgithub.com\u002Faharada54914\u002Fsdd-forge\u002Fblob\u002Fmain\u002Fskills\u002Fadversarial-review\u002FREADME.md",{"path":50,"resourceType":45,"resourceTypeDesc":46,"sizeBytes":51,"sourceHtmlUrl":52},"references\u002Freviewer-prompts.md",9602,"https:\u002F\u002Fgithub.com\u002Faharada54914\u002Fsdd-forge\u002Fblob\u002Fmain\u002Fskills\u002Fadversarial-review\u002Freferences\u002Freviewer-prompts.md",{"path":54,"resourceType":45,"resourceTypeDesc":46,"sizeBytes":55,"sourceHtmlUrl":56},"templates\u002Freport-template.md",4822,"https:\u002F\u002Fgithub.com\u002Faharada54914\u002Fsdd-forge\u002Fblob\u002Fmain\u002Fskills\u002Fadversarial-review\u002Ftemplates\u002Freport-template.md",[58,66,73,80,87,94],{"skillNo":59,"name":60,"nameZh":6,"description":61,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":62,"categories":63,"keywords":64,"publishedAt":65},"01M7DY6PWMGRM0GFNBXAE2RP3K","status","Show a read-only Jira-style six-view SDD dashboard in the initiating parent's Sites or Claude Artifacts, with source freshness and failure visibility.","https:\u002F\u002Fgithub.com\u002Faharada54914\u002Fsdd-forge\u002Fblob\u002Fmain\u002Fplugins\u002Fsdd-ship\u002Fskills\u002Fstatus\u002FSKILL.md",[],[],"2026-11-15 14:19:17",{"skillNo":67,"name":68,"nameZh":6,"description":69,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":70,"categories":71,"keywords":72,"publishedAt":18},"01M75FN16WRENJ2A9CXZCXVKBQ","verify-review-reservation","Recheck the bounded identity-ledger reservation invariant after lock, hash-check or publication changes.","https:\u002F\u002Fgithub.com\u002Faharada54914\u002Fsdd-forge\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fproject-skills\u002Fformal-review-reservation\u002FSKILL.md",[],[],{"skillNo":74,"name":75,"nameZh":6,"description":76,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":77,"categories":78,"keywords":79,"publishedAt":18},"01M75FMW90YZDCB7QNJ8K5KGKG","aws-cdk","Authors, deploys, and troubleshoots AWS infrastructure using CDK with TypeScript or Python. Covers best practices, stack architecture, and construct patterns. Applies when writing CDK constructs, bootstrapping environments, running cdk deploy\u002Fsynth\u002Fdiff, fixing CDK or CloudFormation errors, planning stack structure, importing existing resources, resolving drift, or refactoring stacks without resource replacement.","https:\u002F\u002Fgithub.com\u002Faharada54914\u002Fsdd-forge\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fcloud-orchestration\u002Faws-bundle\u002Fupstream\u002Fplugins\u002Faws-core\u002Fskills\u002Faws-cdk\u002FSKILL.md",[],[],{"skillNo":81,"name":82,"nameZh":6,"description":83,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":84,"categories":85,"keywords":86,"publishedAt":18},"01M75FMSTFNT7EQSZY81T4WRHN","quality-gate","Independently verify an Implementation Complete task with a Default-FAIL verification contract, deterministic checks, and an isolated critical reviewer, then decide whether it is Done.","https:\u002F\u002Fgithub.com\u002Faharada54914\u002Fsdd-forge\u002Fblob\u002Fmain\u002Fspecs\u002Fsdd-domain-multitarget\u002Fverification\u002FT-003\u002Fattempt10-20261016\u002Fbaseline\u002Fplugins\u002Fsdd-quality-loop\u002Fskills\u002Fquality-gate\u002FSKILL.md",[],[],{"skillNo":88,"name":89,"nameZh":6,"description":90,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":91,"categories":92,"keywords":93,"publishedAt":18},"01M75FMPM8FDE4NQ6FFY7KB83Y","ship","Implement approved SDD tasks through the quality gate to Done — the second of the two-command workflow. Run after \u002Fsdd-bootstrap:bootstrap and human task approval.","https:\u002F\u002Fgithub.com\u002Faharada54914\u002Fsdd-forge\u002Fblob\u002Fmain\u002Fspecs\u002Fquality-loop-fixes\u002Fhuman-copy\u002Fplugins\u002Fsdd-ship\u002Fskills\u002Fship\u002FSKILL.md",[],[],{"skillNo":95,"name":89,"nameZh":6,"description":90,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":96,"categories":97,"keywords":98,"publishedAt":18},"01M75FMK3G0EPXW77XK8ED3F2E","https:\u002F\u002Fgithub.com\u002Faharada54914\u002Fsdd-forge\u002Fblob\u002Fmain\u002Fspecs\u002Fmcp-readonly-preflight\u002Fhuman-copy\u002Fplugins\u002Fsdd-ship\u002Fskills\u002Fship\u002FSKILL.md",[],[]]