ai-verify
Use when work has to be verified — "verify the feature", "did we build what was asked", "review this diff", "is it ready to commit", or when the user wants over-engineering findings or a debt ledger. Choose the right verification tier and produce verdicts with evidence, not impre
Install
npx skills add https://github.com/arcasilesgroup/ai-engineering/tree/main/skills/ai-verify
claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install arcasilesgroup-ai-engineering@llmmart
git clone https://github.com/arcasilesgroup/ai-engineering.git
The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole arcasilesgroup/ai-engineering collection as a plugin from our marketplace. Git is the plain clone.
Skill manifest
ai-verify — verification in three tiers (Graph Engineering)
Three ways to verify, depending on who fires them: lanes you run on purpose,
lanes that run themselves mid-workflow, and a chain of review lanes. Every lane is a file in
lanes/: read it when its line below is the one you need. The
chain's router decides which lanes run from the evidence in the diff, never
from its size.
Tier 1 — standalone (you run it on purpose)
- Per-requirement verdict against what was actually asked, with a list of gaps → lanes/verify-feature.md
- Extract the requirements from the conversation into numbered, falsifiable claims → lanes/spec-extract.md
Tier 2 — embedded (fires by itself, mid-workflow)
- Browser verification of the finished feature + blast radius of the routes that depend on the change → lanes/feature-verify.md, with
scripts/blast_radius.ts,scripts/setup_harness.shandreferences/feature-verify/ - A second opinion from a fresh model instance with no visibility into your reasoning → lanes/second-opinion.md, with
scripts/second-opinion.sh
Tier 3 — chain (review lanes picked by evidence)
- Router that picks lanes, estimates cost, and executes → lanes/review-router.md
- Lanes: code-audit · security-audit · a11y-audit · perf-audit · design-check · build-check
- Merge of all lanes into one deduplicated report → lanes/full-review.md; shared lane conventions are in
references/review/CONVENTIONS.md
Setup guide and entry router → VERIFICATION-SETUP-GUIDE.md. The internal eval fixtures apply real bugs and score the reviewer in ai-engineering's own tests.
Lifecycle
Lane: light, standard, full Writes: the verdict, and the open checkpoint's review notes when one exists Read by: CI, the human Dies: with the milestone Next: ai-security when the security trigger fired; ai-write when the public interface changed; ai-visual-recap otherwise
Source: the Graph Engineering verification-skills lineage, generalized to any stack and supplied by the maintainer (rights attested 2026-11-06, see NOTICE.md).
The ai-engineering seam
- The router picks the tier using the source's taxonomy.
- The judge runs at tier
decide; the model comes from the ai-engineeringconfig.tomlpin. Binary checks run at tierverify. - The evals (
tests/fixtures/verify-tiers/evals/scripts/apply-pack.tsandscore.ts) run in ai-engineering's own CI, not in the user's project. - Outputs: the verdict is the report. When a checkpoint is open, the same verdict is written into that checkpoint's review notes. Style follows ai-brainstorm › references/artifact-design.md.
Over-engineering review and debt collection
Use this route for "review code", "code review", "find over-engineering", "check for bloat", or "what can we simplify". It is distinct from judging whether requirements were met: use the verification tiers above for correctness.
Diff review (default)
Review the current PR, branch, or requested diff against the AGENTS.md code style rules. Return ranked findings, one line each: location, problem, fix.
src/utils/parser.ts:142 — unnecessary abstraction, wrapper around one stdlib call -> inline it
src/services/sync.ts:87-94 — generic name hides purpose -> rename it to describe the value
Debt collection
Use /ai-verify --debt or "collect debt", "list deferred shortcuts", or "what did we defer". Scan the codebase for todo: comments and return a structured ledger with file, line, comment, and category (lock, perf, simplification, test, correctness, or other).
For both modes, check findings against the AGENTS.md code style rules: KISS, YAGNI, DRY, early returns, one function/one job, purpose-named values, comments that explain why, and the shortest working solution. Cite the applicable rule in each finding. Do not refactor working code or report a style preference without a concrete improvement.
Files (ai-engineering)
-
lanes
-
a11y-audit.md 8.7 KB
Loading… -
build-check.md 6.6 KB
Loading… -
code-audit.md 7.1 KB
Loading… -
design-check.md 7.6 KB
Loading… -
feature-verify.md 16.3 KB
Loading… -
full-review.md 7.6 KB
Loading… -
perf-audit.md 8 KB
Loading… -
review-router.md 9.4 KB
Loading… -
second-opinion.md 9.6 KB
Loading… -
security-audit.md 9.4 KB
Loading… -
spec-extract.md 9.7 KB
Loading… -
verify-feature.md 12.1 KB
Loading…
-
-
references
-
feature-verify
-
harness.md 8.2 KB
Loading… -
puppeteer-patterns.md 14.9 KB
Loading…
-
-
review
-
CONVENTIONS.md 7.8 KB
Loading…
-
-
-
scripts
-
blast_radius.ts 16.5 KB
Loading… -
second-opinion.sh 2.1 KB
Loading… -
setup_harness.sh 11.9 KB
Loading…
-
-
SKILL.md 4.7 KB
--- name: ai-verify description: >- Use when work has to be verified — "verify the feature", "did we build what was asked", "review this diff", "is it ready to commit", or when the user wants over-engineering findings or a debt ledger. Choose the right verification tier and produce verdicts with evidence, not impressions. Not for milestone-declared security audits — use /ai-security. Not for diagnosing a failure — use /ai-debug. license: Apache-2.0 --- # ai-verify — verification in three tiers (Graph Engineering) Three ways to verify, depending on who fires them: lanes you run on purpose, lanes that run themselves mid-workflow, and a chain of review lanes. Every lane is a file in `lanes/`: read it when its line below is the one you need. The chain's router decides which lanes run from the evidence in the diff, never from its size. ### Tier 1 — standalone (you run it on purpose) - Per-requirement verdict against what was actually asked, with a list of gaps → [lanes/verify-feature.md](lanes/verify-feature.md) - Extract the requirements from the conversation into numbered, falsifiable claims → [lanes/spec-extract.md](lanes/spec-extract.md) ### Tier 2 — embedded (fires by itself, mid-workflow) - Browser verification of the finished feature + blast radius of the routes that depend on the change → [lanes/feature-verify.md](lanes/feature-verify.md), with `scripts/blast_radius.ts`, `scripts/setup_harness.sh` and `references/feature-verify/` - A second opinion from a fresh model instance with no visibility into your reasoning → [lanes/second-opinion.md](lanes/second-opinion.md), with `scripts/second-opinion.sh` ### Tier 3 — chain (review lanes picked by evidence) - Router that picks lanes, estimates cost, and executes → [lanes/review-router.md](lanes/review-router.md) - Lanes: [code-audit](lanes/code-audit.md) · [security-audit](lanes/security-audit.md) · [a11y-audit](lanes/a11y-audit.md) · [perf-audit](lanes/perf-audit.md) · [design-check](lanes/design-check.md) · [build-check](lanes/build-check.md) - Merge of all lanes into one deduplicated report → [lanes/full-review.md](lanes/full-review.md); shared lane conventions are in `references/review/CONVENTIONS.md` Setup guide and entry router → [VERIFICATION-SETUP-GUIDE.md](VERIFICATION-SETUP-GUIDE.md). The internal eval fixtures apply real bugs and score the reviewer in ai-engineering's own tests. ## Lifecycle Lane: light, standard, full Writes: the verdict, and the open checkpoint's review notes when one exists Read by: CI, the human Dies: with the milestone Next: ai-security when the security trigger fired; ai-write when the public interface changed; ai-visual-recap otherwise Source: the Graph Engineering verification-skills lineage, generalized to any stack and supplied by the maintainer (rights attested 2026-11-06, see NOTICE.md). ## The ai-engineering seam 1. The router picks the tier using the source's taxonomy. 2. The judge runs at tier `decide`; the model comes from the ai-engineering `config.toml` pin. Binary checks run at tier `verify`. 3. The evals (`tests/fixtures/verify-tiers/evals/scripts/apply-pack.ts` and `score.ts`) run in ai-engineering's own CI, not in the user's project. 4. Outputs: the verdict is the report. When a checkpoint is open, the same verdict is written into that checkpoint's review notes. Style follows [ai-brainstorm › references/artifact-design.md](../ai-brainstorm/references/artifact-design.md). ## Over-engineering review and debt collection Use this route for "review code", "code review", "find over-engineering", "check for bloat", or "what can we simplify". It is distinct from judging whether requirements were met: use the verification tiers above for correctness. ### Diff review (default) Review the current PR, branch, or requested diff against the AGENTS.md code style rules. Return ranked findings, one line each: location, problem, fix. ```text src/utils/parser.ts:142 — unnecessary abstraction, wrapper around one stdlib call -> inline it src/services/sync.ts:87-94 — generic name hides purpose -> rename it to describe the value ``` ### Debt collection Use `/ai-verify --debt` or "collect debt", "list deferred shortcuts", or "what did we defer". Scan the codebase for `todo:` comments and return a structured ledger with file, line, comment, and category (`lock`, `perf`, `simplification`, `test`, `correctness`, or `other`). For both modes, check findings against the AGENTS.md code style rules: KISS, YAGNI, DRY, early returns, one function/one job, purpose-named values, comments that explain why, and the shortest working solution. Cite the applicable rule in each finding. Do not refactor working code or report a style preference without a concrete improvement. -
VERIFICATION-SETUP-GUIDE.md 47.4 KB
Loading…
Comments (0)
Sign in to join the conversation.
Reviews (0)
No reviews yet.
No comments yet.