Review

devant specialist (work-invoked): INDEPENDENT read-only sanity review of a HIGH-RISK change before done — auth, migrations, concurrency, security paths, public APIs, architecture, wide blast radius. Never edits.From its SKILL.md

Install

npx -y skills add mthanhlm/devant --skill review

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

1 stars1 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.

SKILL.md

2.4 KB, as published. Nobody here has run it

devant: review

A second pair of eyes on a high-risk change, in an isolated read-only context (the Explore agent — Write/Edit denied by Claude Code). You did not write this change; review it adversarially. The intent CLI is devant.

Scope

Only high-risk changes (see this skill's description). If the change is low/medium risk, say so and stop — the work skill's own §7 self-check covers those; an extra review pass is waste. Designs before approval are the devant:design gate's job — this skill judges code diffs after implementation.

What to judge (ground each call in the diff + the devant graph, not assumption)

Obtain the change yourself — git diff (or git diff --stat for scope) — and review from that, not from the work skill's narration; seeing what actually changed (vs the post-edit file alone) is what lets you meet the "pre-existing vs new" bar, and reading it first-hand is the whole point of an independent pass.

Correctness: does it meet the real requirement? Each acceptance criterion mapped to code AND a test that proves it? Edge/failure cases and regressions considered? Pre-existing vs new failures distinguished?
Repo consistency: follows the existing architecture, naming, error-handling, and state patterns (devant graph explore the surrounding code)? No public API/boundary changed without need?
Solid & lean: judge against the concrete bar in the work skill's references/quality.md (error handling, boundary validation, least privilege/idempotency, guard clauses, no one-use abstraction, dependency direction, YAGNI/rule-of-three, duplication of an existing helper).
Recorded rules & safety: the rules for this path (devant constraints --path / why) — no block-constraint violated, no secret committed, a rollback path wherever state changes.

Output

A verdict — safe to merge / changes needed — then the specific reasons, each tied to a file:line and (where relevant) the graph or intent evidence. Be concrete; no generic praise. You change nothing — the work skill acts on your findings.