AgentStack
Browse Sign in
Browse Why AgentStack Sell Docs
Sign in
SKILL verified MIT Self-run

Anchor Vet

skill-lynxlangya-techne-anchor-vet · by lynxlangya

Evidence-gated diff review for PRs, branches, commit ranges, staged code changes, and merge readiness checks. Use when an AI reviewer must review a concrete git-anchored diff, judge whether code is safe to merge, inspect a PR URL or branch, cross-examine PR or commit claims, account blast radius, or render approve/request-changes/blocked verdicts. Do not use for design-only review, whole-codebase…

— No reviews yet
0 installs
24 views
0.0% view→install

Install

$ agentstack add skill-lynxlangya-techne-anchor-vet

✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.

Security review

✓ Passed

No issues found. Passed automated security review. · v0.1.0 How review works →

  • ✓ Prompt-injection patterns
  • ✓ Secret / credential exfiltration
  • ✓ Dangerous shell & filesystem operations
  • ✓ Untrusted network calls
  • ✓ Known-malicious package signatures

What it can access

  • ✓ Network access No
  • ✓ Filesystem access No
  • ✓ Shell / process execution No
  • ✓ Environment & secrets No
  • ✓ Dynamic code execution No

From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.

View the full security report →

Verified badge

Passed review? Show it. Paste this badge into your README, it links to the public security report.

AgentStack Verified badge Links to your public security report.
[![AgentStack Verified](https://agentstack.voostack.com/badges/verified.svg)](https://agentstack.voostack.com/security/report/skill-lynxlangya-techne-anchor-vet)

Reliability & compatibility

✓ Security review passed
0 installs to date
— no reviews yet
● 3mo ago

Declared compatibility

Claude CodeClaude Desktop

Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.

Preview Execution monitoring

We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.

How agent discovery & health will work →
Are you the author of Anchor Vet? Claim this listing to set pricing, connect Stripe payouts, and keep 70% of every sale.
Sign up to claim

About

anchor-vet

Force the skipped move in code review: prove the reviewed diff's scope, blast radius, claims, findings, and verdict are anchored to git evidence.

Trigger Check

Use this skill only when there is a concrete code change to judge: a PR, branch, commit range, or staged review target.

Boundary test: can you name a base...head pair of git states whose difference is the artifact under judgment? If yes, use anchor-vet. If no, get the branch or ref first; do not review pasted diffs or prose descriptions.

Do not use anchor-vet for design feedback without a diff, whole-codebase audits, feature implementation, bug fixing, formatting-only tasks, or document review. If the user asks you to fix findings too, render the review verdict first; fixes are a separate task, and behavioral fixes route to anchor-repro.

Forced Procedure

  1. Anchor the scope. Check out the reviewed head locally. Capture external

claims before init: for PRs, save the title/body from gh pr view --json title,body --jq '.title + "\n\n" + (.body // "")' to a claims file. Run: python3 skills/anchor-vet/scripts/vet_gate.py init --project --review --base --head --claims-file or, only when there are genuinely no external claims, --no-claims.

  1. Read every hunk. Inspect the full diff and every hunk in scope.json.

If the diff is too large to read honestly, stop and propose a split.

  1. Walk the blast radius. For each candidate symbol, read the references

found by the gate. Record examined refs with effect and a one-line note. Examined means you opened/read the reference and judged how the change affects it.

  1. Account weak or symbolless hunks. For weak/symbolless hunks, use a

verified enclosingUnit when a named unit exists, or file-level with a reason only for genuinely unit-less code/config/prose.

  1. Cross-examine claims. Disposition every anchored claim id from

scope.json: verified, contradicted, not-verifiable-from-diff, or non-claim. Verified/contradicted claims need citations.

  1. Hunt findings with severity honesty. Use only blocking, concern, and

nit. Cite findings. A blocking finding needs R2 cited evidence or an R3 repro probe. When a behavioral assertion is cheap to demonstrate, record a failing anchor-repro ledger entry against the reviewed head and cite it with entrySha256.

  1. Write and check review.json. Run:

python3 skills/anchor-vet/scripts/vet_gate.py check --project --review . Fix check failures by doing the missing review work, not by padding JSON.

  1. Render the verdict through the gate. Run:

python3 skills/anchor-vet/scripts/vet_gate.py close --project --review --verdict approve|request-changes|blocked. Report the verdict, cite verdict.json, and name each finding's evidence rung.

Script Contract

Artifacts are written under the target project:

.techne/review//
  scope.json    # computed by init
  review.json   # authored by the reviewer
  report.json   # computed by check
  verdict.json  # computed by close

Generated .techne/ output belongs to target projects. Do not commit it to this repository.

Review skeleton:

{
  "schema": "techne.vet/1",
  "symbols": [
    {
      "id": "s-...",
      "symbol": "changedName",
      "disposition": "reviewed",
      "refs": [
        {
          "file": "src/caller.py",
          "line": 42,
          "effect": "unaffected",
          "note": "caller passes the same contract"
        }
      ]
    }
  ],
  "hunks": [],
  "claims": [
    {
      "ids": ["c-..."],
      "disposition": "verified",
      "citations": [{"file": "src/change.py", "line": 10}]
    }
  ],
  "findings": [],
  "testsAcknowledgment": "No tests changed; the diff is docs-only or existing coverage is enough because ..."
}

Use reference.md for the full JSON schema, rubrics, exclusions, weak spots, and probe guidance.

Stop Conditions

  • Stop if no git-anchored base...head diff is available.
  • Stop if init refuses because of dirty worktree, missing merge base, or head

mismatch; fix the setup or record blocked.

  • Stop if the review is too large to read in full; propose a split instead of

sampling silently.

  • Stop before approve if refs, hunks, claims, or test acknowledgment remain

unaccounted.

  • If head moves mid-review, do not fight the tool. Close the old slug as

blocked --reason ... or leave it, then start a new slug for the new head.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

Install and usage instructions live in the source repository linked above.

Reviews

No reviews yet, be the first.

Versions

  • v0.1.0 Imported from the upstream source.