| name | loom-code-review |
| description | Run multi-axis implementation review through Loom critique. Use before merge, after feature or bug-fix implementation, when receiving review feedback, when reviewing AI/human code, or when correctness, readability, architecture, security, performance, tests, or evidence need pressure-testing. |
| compatibility | Markdown-native, script-free Loom protocol. |
| metadata | {"skill_kind":"workflow"} |
loom-code-review
Code review is adversarial evidence and implementation judgment, not politeness or rubber-stamping.
This playbook adapts peer code-review request and reception workflows into Loom's
critique, ticket, evidence, and ship layers.
Core Dependency
This playbook requires loom-core. If using-loom and the core owner-layer
skills are not installed or preloaded, stop and load/install loom-core instead
of treating this playbook as a substitute for Loom doctrine or record grammar.
What This Workflow Coordinates
- five-axis review: correctness, readability, architecture, security, performance
- test and evidence review before acceptance
- review packet/context preparation for humans or agents
- external review feedback classification and response
- ticket-owned finding disposition and follow-up routing
What This Workflow Does Not Own
- critique verdict records; use core critique
- ticket acceptance, finding disposition, or closure; use tickets
- evidence artifacts; use evidence
- Git provenance; use
loom-git
- PR or release wording; use
loom-ship
Use This Skill When
- reviewing a diff before merge or acceptance
- an AI agent or human produced code that needs independent scrutiny
- a bug fix needs both implementation and regression-test review
- receiving review comments that may be correct, unclear, optional, or wrong
- a change is large enough that tests passing is not sufficient evidence
Do Not Use This Skill When
- the request is only to read current code and explain it; use codemap or research
- no implementation or reviewable artifact exists yet
- the target is a Loom record rather than code; use core critique directly
- the next step is only packaging already-reviewed work; use
loom-ship
Default Procedure
- Understand intent before diff details: spec, ticket, acceptance IDs, plan,
evidence, risk class, and claimed behavior.
- Review tests first. Ask whether they prove behavior, cover edge/error cases, and
would fail for the likely regression.
- Review implementation across correctness, readability, architecture, security,
and performance.
- Check change size and concern mixing. Ask for split work when reviewability is poor.
- Verify the verification story: commands, browser evidence, scans, screenshots,
benchmarks, skipped checks, and stale outputs.
- Record durable findings in critique when they should persist. Classify severity,
confidence, evidence reviewed, residual risk, and required follow-up.
- Route every open medium/high finding to ticket-owned disposition before closure.
- When receiving feedback, evaluate it technically before implementing. Clarify
unclear items, push back on incorrect or out-of-scope requests, and implement
valid fixes one at a time with verification.
- Use
loom-ship only after review disposition and evidence are truthful.
Finding Severity And Disposition
Use core critique severity in saved critique findings: low, medium, or
high. Put action semantics in the finding title, impact, or required follow-up;
do not replace core severity with peer labels such as Critical, Important,
Suggestion, Nit, or FYI.
high: likely blocks acceptance unless the ticket records resolved,
accepted_risk, superseded, or converted_to_follow_up disposition.
medium: material risk or quality gap that needs ticket-owned disposition
before closure.
low: useful concern or polish issue; record it when it should persist, but it
does not block closure by default.
For open medium/high findings, use only the ticket-owned finding dispositions
from core: resolved, accepted_risk, superseded, or
converted_to_follow_up.
Common Rationalizations
- Rationalization: "Tests pass, so review can be quick."
Reality: Tests do not cover architecture, security, readability, performance, or evidence sufficiency.
- Rationalization: "It's AI-generated but looks plausible."
Reality: AI code needs more scrutiny because it can be confidently wrong.
- Rationalization: "Reviewer feedback is an order."
Reality: Feedback is a claim to evaluate against codebase truth, scope, and owner records.
- Rationalization: "We can clean it up later."
Reality: Deferred cleanup needs ticket disposition; otherwise it usually disappears.
Red Flags
LGTM without evidence of axes reviewed
- no review of tests or evidence
- required and optional comments are indistinguishable
- large mixed feature/refactor/config changes reviewed as one blob
- security-sensitive diff lacks security review
- external feedback implemented before verification or clarification
- medium/high findings lack ticket disposition
Verification
Done Means
- critique and ticket records tell the truth about review outcome
- evidence supports or limits the acceptance claim
- required findings are
resolved, accepted_risk, superseded, or
converted_to_follow_up
- optional feedback is not treated as hidden mandatory scope
- shipping summaries can mirror owner truth without inventing it
Read In This Order
Read immediately for code review:
references/review-and-feedback-loop.md for five-axis review, review packet
context, core severity/disposition usage, external feedback handling, and
dependency review.
- the core
loom-critique skill for durable findings and verdicts.
- the core
loom-tickets skill for finding disposition and closure gate.
- the core
loom-evidence skill for verification story checks.
Then read conditionally:
skills/loom-security/SKILL.md or skills/loom-performance/SKILL.md for specialized risk.
skills/loom-git/SKILL.md when base/head, diff, branch, or PR provenance matters.
skills/loom-ship/SKILL.md when packaging review outcome for PR or release.