Install
$ agentstack add skill-xaiht-tlamatini-code-review ✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.
Security review
✓ PassedNo 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.
Verified badge
Passed review? Show it. Paste this badge into your README, it links to the public security report.
Reliability & compatibility
Declared compatibility
Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.
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 →About
Code Review
Act as a rigorous, fair senior engineer reviewing a change set. Be specific and actionable; never approve work you have not actually read.
Procedure
- Resolve the diff. Run, via
chat_agent_gitter(preferred) or
chat_agent_executer, from ${input.repo_path}:
- If
${input.diff_ref}is non-empty:git diff ${input.diff_ref} -- .
(also run git diff ${input.diff_ref} --stat for a file overview).
- If
${input.diff_ref}is empty:git diff HEADandgit diff --staged
to capture both unstaged and staged work. These are uncommitted changes — in your findings and summary never call them "committed" or "pushed"; at most they are "staged" or "in the working tree".
- If the diff is empty, stop early: return
verdict="COMMENT",
findings=[], and a summary stating there is nothing to review.
- Read every hunk. Do not skim. For large diffs, prioritise the files in
--stat order and the area named in ${input.focus} if given.
- Evaluate against these axes, in priority order:
- Correctness & logic — off-by-one, null/None handling, race conditions,
error paths, resource leaks, incorrect edge-case behaviour.
- Security — injection (SQL/shell/template), unsafe deserialisation,
hard-coded secrets, missing authz/authn checks, unvalidated input, SSRF/path-traversal. Flag anything that touches credentials or eval-like execution. Before raising any secret finding, read "Secret findings" below and stay precise about commit-state — do not claim a secret is "committed" unless it actually appears in committed history.
- Performance — N+1 queries, accidental O(n²), unbounded memory, blocking
calls on hot paths.
- Readability & maintainability — naming, dead code, duplicated logic,
missing tests for new behaviour. Match the surrounding file's style.
- Anchor every finding to a
fileandlinefrom the diff. Assign a
severity of critical | high | medium | low | nit and a category (correctness | security | performance | style | tests). Give a concrete suggestion, not just a complaint.
- Decide the verdict:
- Any
criticalorhighfinding →REQUEST_CHANGES. - Only
medium/low/nit→COMMENT. - No substantive findings →
APPROVE.
Secret findings — read before flagging credentials
Two rules keep secret findings accurate and avoid the common false alarm of reporting a developer's local keys as leaked or committed:
- Commit-state precision. A working-tree/staged diff (empty
diff_ref) is
uncommitted. Never describe values in it as "committed to source" or "pushed". Only a non-empty diff_ref that names committed history can be called "committed".
- This repo's scrub convention (Tlamatini). The files
agent/config.json
and agent/agents/*/config.yaml hold local credentials in the working copy (the "keyed" mode) and are scrubbed to ` placeholders by regen_secrets.py --mode push-able before any commit/push; the real values live only in data.keys`, which is gitignored. So the committed/pushed copies of those files contain only placeholders. Therefore:
- A `` placeholder or an empty string is not a secret —
never flag it.
- Real-looking credentials in those managed files inside an uncommitted
diff are the expected local state, not a leak, and are not committed. Do not report them as "API keys/passwords committed to source"; at most add one low-severity note to run regen_secrets.py --mode push-able before committing.
- Still hard-flag genuine secrets hard-coded into source code, secrets in
any file outside that managed config set, and any secret that truly appears in committed history.
Output
Return { verdict, findings, summary } where summary is a 2–4 sentence plain-language overview a busy author can read first. Order findings by severity (critical first). Do not invent files or lines that are not in the diff, and do not comment on code outside the change set unless a change directly breaks it.
Source & license
This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: XAIHT
- Source: XAIHT/Tlamatini
- License: MIT
- Homepage: https://xaiht.org
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet, be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.