codex-review
Skill45 starsChanged 52 days agoMITView on GitHub
Run a bounded read-only implementation review with Codex as a native background subagent from Codex, or through the non-interactive Codex CLI from another harness. Never create a Herdr pane, tab, workspace, or agent for a reviewer.
91 other projects use a version of this.
Adopted by
Most recently taken up by
- p-chan/agent-skillstheir own version, first seen Nov 8, 2026
- p-chan/skillstheir own version, first seen Nov 8, 2026
- Smirnov-Labs/claude-skillstheir own version, first seen Nov 8, 2026
- juriwiens/skillstheir own version, first seen Nov 8, 2026
- ryanportfolio/winuse-mcptheir own version, first seen Nov 8, 2026
- fab-codes/agent-workflow-templatetheir own version, first seen Nov 8, 2026
Used with
Projects with it also use addskill, apply-firmware and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
ryanportfolioSkilladdskill
Create, import, update, or install repo or personal skills: runtime ownership, resources, discovery checks.
Also in 6 projects that use this
ryanportfolioSkillapply-firmware
Bring current folder (plain folder, existing repo, old or partial firmware copy) up to full Harness Firmware in place. Mirror external repo → adopt-repo.
Also in 6 projects that use this
- +10 more from ryanportfolio
Other versions91 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 91
- ClickHouse/ClickHouse★ 51ksame name
- chaseai-yt/claudex-loop★ 3.2ksame name
- morethanwords/tweb★ 2.7ksame name
- alinaqi/maggy★ 711same name
- alinaqi/claude-bootstrap★ 708same name
- ThibautBaissac/rails_ai_agents★ 665same name
- ai-analyst-lab/ai-analyst★ 304same name
- Sateezg/codex-bridge★ 248same name
- sd0xdev/sd0x-dev-flow★ 192same name
- sd0xdev/sd0x-harness★ 192same name
81 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository11 other files it gives its agents.
- AGENTS.mdAGENTS.md
- tests/AGENTS.mdAGENTS.md
- xtask/AGENTS.mdAGENTS.md
- architecture-reviewSkill
- claude-code-reviewSkill
- implement-in-worktreeSkill
- proqi-screenshotSkill
- releaseSkill
- walkthroughSkill
- proqi-debugSkill
- proqiSkill
The file
We split this skill into the rules it gives an agent
9 rules. 0 are common, 6 are distinctive, and 3 are very specific to this project.
0
Common
Most of the 79 edited versions we read keep these.
6
Distinctive
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 8
Run one independent Codex review without turning the reviewer into an interactive terminal participant.
- Very specific: line 19
Treat uncommitted files as a moving snapshot.
- Very specific: line 26
Never create or split a Herdr pane, tab, workspace, or worktree for it.
- Distinctive: line 35
When the invoking harness is Codex, use exactly one native Codex subagent through the harness's subagent or collaboration mechanism.
- Very specific: line 37
Do not run the review in the primary agent.
- Distinctive: line 43
Only when the invoking harness is not Codex, launch one ephemeral, non-interactive Codex CLI process through that harness's managed process …
- Distinctive: line 45
Use the exact requested model when one is given.
- Distinctive: line 81
Wait for the native subagent or managed CLI process only when its result is needed.
- Distinctive: line 86
Verify every finding against the current source.
“Common” is measured: kept by at least 6 in 10 of the 79 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Codex Review
Run one independent Codex review without turning the reviewer into an interactive terminal participant. The implementation owner remains the only interactive coding agent in its Herdr worktree.
Establish the review snapshot
- Resolve the repository, worktree, pull request, commit, diff, or files in
scope. Read every applicable
AGENTS.mdand relevant product or architecture contract before composing the assignment. - Record the absolute working directory, branch, exact
HEAD, requested base, andgit status --short. Preserve unrelated changes. - Treat uncommitted files as a moving snapshot. Compare
HEADand status again after the review. A final stable-head verdict requires a fresh review if the reviewed files changed during the pass.
Reviewer execution is never interactive
A reviewer must run as a non-interactive background worker relative to Herdr.
Never create or split a Herdr pane, tab, workspace, or worktree for it. Never
use herdr agent start, herdr agent prompt, or another terminal multiplexer
to host a reviewer. The reviewer inspects the implementation owner's existing
worktree.
Determine the invoking harness before dispatch. A Codex agent hosted inside Herdr is still running in the Codex harness. Herdr is only the terminal host and does not change the harness identity.
When the invoking harness is Codex, use exactly one native Codex subagent
through the harness's subagent or collaboration mechanism. This is mandatory.
Do not run the review in the primary agent. Never launch the codex executable,
including codex exec or codex exec review, from a Codex agent. If that Codex
agent cannot spawn a native subagent, report the blocker to its parent session.
The parent Codex session may dispatch the native reviewer into the same existing
worktree. Do not fall back to the CLI, a Herdr reviewer, or primary-agent review.
Only when the invoking harness is not Codex, launch one ephemeral,
non-interactive Codex CLI process through that harness's managed process runner.
Use the exact requested model when one is given. Otherwise retain the configured
Codex default. Use the review worktree as --cd, select the read-only sandbox,
and capture complete standard output and standard error in one collision-free
file under /private/tmp. A representative shape is:
codex exec --ephemeral --sandbox read-only --cd REPOSITORY_ROOT \
--color never REVIEW_PROMPT > RESULT_FILE 2>&1The process may continue after the runner initially yields, but it must remain
owned by that managed process session until exit. Do not use shell &, nohup,
an interactive Codex session, or a dedicated terminal or Herdr pane.
Give a strictly read-only assignment
The review prompt must include:
- the exact scope, base, and recorded
HEAD; - relevant contracts and acceptance criteria;
- an instruction to inspect and report only;
- a prohibition on editing, creating, deleting, renaming, formatting, or otherwise mutating tracked or untracked repository files;
- a prohibition on commits, pushes, pull-request changes, comments, reviews, process control, and other external mutations;
- severity-ordered findings with concrete file and line evidence;
- a request to report no findings plainly when none are supported.
The reviewer may run read-only inspection commands and non-destructive tests when they materially validate a finding. Tests may create ordinary ignored build or temporary output, but the reviewer must not update snapshots, fixtures, lockfiles, generated sources, or repository state. It must never weaken a test or gate.
Verify the result
- Wait for the native subagent or managed CLI process only when its result is needed. Record its completion status. For a CLI review, read the complete result file even when the process exits nonzero.
- Re-record
HEADandgit status --shortand classify the result as stable or moving. - Verify every finding against the current source. Treat it as a hypothesis, not authority. Classify it as confirmed, partially confirmed, rejected with evidence, already fixed during a moving review, or requiring a product decision.
- The reviewer never implements fixes. The implementation owner may fix confirmed findings when its assignment authorizes that work. A later pass must use a fresh subagent or fresh CLI process.
Report the reviewed scope, exact revisions, execution route, moving-snapshot status, process outcome, and verified findings. Keep a CLI result file until its findings are incorporated. Inspect its exact path before removing only that owned artifact.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

