agentsclimarketplace

Review work

Skill hungv47/meta-skills/skills/meta/review-work

Turn your coding agent into a full product team — composable skills across research, marketing, product, and process. One install, every editor (Claude Code, Cursor, Codex, Windsurf, Gemini CLI, VS Code).

Install
npx -y skills add hungv47/meta-skills --skill review-work

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 13 stars13 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Independent post-implementation review — an agent with no sunk-cost bias checks just-written code or an artifact against its requirements, then resolves the issues found (max 2 rounds). Use to verify a change before shipping, get a second opinion, or check what you missed; auto-triggers for security-sensitive and data-mutation code. Not for code refactoring (use clean-code) or decision analysis (use debate-agents).

SKILL.md

6.5 KB, ~1.4k tokens by cl100k_base, as published. Nobody here has run it

Review Chain — Fresh-Eyes Post-Implementation Quality

Independent reviewer + resolver loop for code, artifacts, plans, or shipped work. Reviewer has no implementation context; resolver synthesizes. Routing in routing.yaml; methodology + "no critic on critic" in references/playbook.md.

Core: What would a senior reviewer with no sunk-cost bias catch?

Critical Gates — load first

  1. Reviewer has NO access to implementation reasoning — only the output + requirements.
  2. Resolver sees BOTH original + review — synthesizes, not patches.
  3. Max 2 loops. Not clean after 2 cycles → flag user; may be a design problem review can't fix.
  4. Auto-trigger for critical code — security, auth, crypto, data mutations, money, PII. Don't wait to be asked.
  5. Quality feedback — repeated misses, critic overrides, post-humanmaxxing regressions feed _shared/quality-feedback-protocol.md, not just this report.
  6. Noise-filter before report-write. Every finding passes noise-filter.md: Layer 1 (real-vs-fake per procedures/reviewer.md § Verification rules) then Layer 2 (Accepted / Rejected / Deferred — three report subsections). Accepted clears fix-then-rerun before Verified.

Before Starting

Apply references/_shared/before-starting-check.md. Then:

  • Mode (references/_shared/mode-resolver.md): default standard; auto-escalate to deep for auth/sessions/access-control/payments/financial-data/migrations/bulk-mutations/PII OR diff >500 lines; auto-downgrade to fast for typos/log lines/config tweaks.
  • references/_shared/execution-policy.md — session execution profile (single-vs-multi)
  • Read docs/forsvn/artifacts/meta/specs/*.md + tasks.md if present — enables scope-drift detection per procedures/scope-drift.md.
  • Read docs/forsvn/artifacts/meta/records/learned-rules.md — append to reviewer CONTEXT.

Mode map: fast = generalist, skip resolver if PASS; standard = generalist + resolver loop; deep = 3 specialists parallel OR critic-consensus (non-code).

Artifact Contract

  • Path: docs/forsvn/artifacts/meta/records/[YYYY-MM-DD]-fresh-eyes-<slug>.md (dated, immutable per-run)
  • Lifecycle: snapshot — accumulates, never overwritten
  • Frontmatter: skill, produced_by, version, date, status, mode, rounds, verdict (PASS/FIXED/CRITICAL), provenance. Template: references/report-template.md
  • Sections: Verdict · Issues Found · Input Quality · Scope Drift · Simplifications · Changes Made · Self-Regulation Gate · Reviewer Summary · Resolver Notes · Specialist Verdicts (deep) · Critic Disagreements (critic-consensus)
  • Consumed by: operator audit; future fresh-eyes runs; commit/PR creation (PASS gate)
  • Eval workspace: none — fresh-eyes IS the eval mechanism.

Pre-Dispatch + Execution

  • Pre-dispatch + Warm/Cold Start: references/_shared/pre-dispatch-protocol.md, references/procedures/pre-dispatch.md.
  • Reviewer + resolver built per-use from procedures/reviewer.md, resolver.md — no static agents dir.
  • 8-step loop (identify → spawn → evaluate → resolve → self-regulation gate → loop → report → deliver) + Configuration: procedures/execution.md.

Auto-trigger rules

Run proactively for: security-sensitive code (auth, crypto, access control); data-mutation code (migrations, bulk updates, deletes); complex/uncertain implementations; money or PII.

Do NOT for: trivial changes (typos, config, log lines), "just do it quick" code, read-only ops.

Anti-Patterns + Edge Cases

Read references/anti-patterns.md before: stacking a critic on the reviewer (banned), giving reviewer implementation reasoning, skipping resolver, auto-applying past self-regulation gate, >2 loops, nit-padding. Edge cases (hallucinations, regressions, oversized code, agent failures, architecture reviews) live there too.

Durable Rules (protected)

<!-- SLOW_UPDATE_START --> <!-- No pinned rules yet. Populate via the slow-update workflow (see references/slow-update-fence.md). Each pinned rule must (a) be procedural not instance-specific, (b) be earned from a regression or critic-flagged failure, (c) cite the artifact / decision record that justified pinning. --> <!-- SLOW_UPDATE_END -->

Completion Status

  • DONE — findings resolved or accepted; PASS gate met.
  • DONE_WITH_CONCERNS — non-blocking issues flagged; report names deferred + why.
  • BLOCKED — critical issue (security/data-loss/broken contract) OR self-regulation gate fired; resolver did NOT apply; operator judgment required.
  • NEEDS_CONTEXT — requirements unclear; missing spec/intent/acceptance criteria.

References

What ships with it: 27 files

224.7 KB alongside SKILL.md, 7 of them executable

references/

scripts/

Gives 0 of the 12 instructions most quality gates skills give in ~1.4k tokens

Counted across 1,195 of the 2,094 authors here whose files we hold, read 2026-08-07

  • Read the output and check the exit codein 54 of 1195, across 14 files
  • Verify requirements using a line-by-line checklistin 53 of 1195, across 12 files
  • Identify the verification command proving the claimin 51 of 1195, across 12 files
  • Run the full verification commandin 50 of 1195, across 11 files
  • Verify output confirms the claimin 49 of 1195, across 12 files
  • Check version control diff after agent delegationin 46 of 1195, across 6 files
  • State claim with evidencein 44 of 1195, across 4 files
  • Run the test suitein 33 of 1195, across 26 files
  • Keep state in memory by defaultin 27 of 1195, across 6 files
  • Make prototype runnable with one commandin 26 of 1195, across 5 files
  • Produce a verification reportin 25 of 1195, across 14 files
  • Detect the package manager from lockfilesin 24 of 1195, across 5 files

Said here and by no other author read

  • restrict reviewer access to output and requirements only
  • give resolver both original output and review findings
  • stop after two unresolved review loops
  • escalate to deep mode for critical code
  • read project specifications to enable scope-drift detection
  • run proactively for security-sensitive and data-mutation code

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

Keep looking

Skills are one crate of 327,132. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.