agentsclimarketplace

Codex changelog

Skill zeikar/hyperclaude/.claude/skills/codex-changelog

Claude Code plugin: Claude builds, Codex critiques — a gated research → plan → review → ship pipeline, with autonomous multi-agent revise loops that self-converge.

Install
npx -y skills add zeikar/hyperclaude --skill codex-changelog

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

  • 3 stars3 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.

What its author says it does

Copied from the file, not written here

This skill should be used when the user asks to check for new Codex CLI (codex-cli) releases, "what changed in Codex", "any Codex updates", "check codex releases", "diff the codex version", or wants Codex CLI releases reviewed for changes relevant to THIS repo's Codex bridge. Reads a locally-stored last-checked codex-cli stable version, lists GitHub releases for openai/codex, reports changes relevant to the plugin's Codex integration (bridge argv, sandbox, exec/resume, JSONL output), and offers to advance the stored version.

SKILL.md

6.9 KB, as published. Nobody here has run it

codex-changelog — Codex CLI release watch (hyperclaude bridge relevance)

Repo-local dev tool (NOT a plugin skill). Compare a stored "last-checked" codex-cli STABLE version against the latest stable release, mine the release notes in between, and report only what touches this repo's Codex bridge — then optionally advance the stored version so the next run diffs from here.

The value is not "summarize Codex releases" — it is mapping generic Codex CLI changes onto the specific surface of the hyperclaude bridge (the only Codex-spawning code in the plugin) and verifying impact against the real bridge files.

Source (differs from cc-changelog)

Codex's CHANGELOG.md only points at GitHub Releases — it is NOT a per-version file. So the authoritative source is the GitHub Releases of openai/codex, read via gh (already authed in this repo):

  • gh release list --repo openai/codex --limit <N> — tags look like rust-v0.142.5 (stable) and rust-v0.143.0-alpha.NN (pre-release). The row marked Latest is the current stable.
  • gh release view <tag> --repo openai/codex — the per-release notes.

Step 2's get-releases.mjs helper wraps both calls (list → filter to new stables → per-new-stable view), so you don't invoke gh by hand in the normal path.

Fallback if gh is unavailable: WebFetch https://github.com/openai/codex/releases.

Track STABLE, note alphas. The plugin runs against the installed stable codex, so the version diff gates on stable→stable. The alpha stream (-alpha.NN, many per week) is NOT tracked per-release — but DO surface a one-line "N newer alphas exist (latest rust-v0.143.0-alpha.NN)" so upcoming changes are visible without firehose noise.

State file

.claude/skills/codex-changelog/.last-checked-version (sibling of this file, gitignored): a single line with the last-reviewed codex-cli STABLE version, e.g. 0.142.5. It is the diff baseline; it advances only in Step 6.

Procedure

Step 1 — Baseline

The diff baseline is .claude/skills/codex-changelog/.last-checked-version (one stable X.Y.Z line) — call it STORED. Step 2's helper reads it automatically; you don't Read it here. If it's missing/empty the helper exits 2 — then ask which stable version to baseline (or read the installed one via codex --version) and pass it as the argument.

Step 2 — List releases + collect new stable notes

Run the local helper — it lists openai/codex releases via gh, finds the latest stable, and prints the notes of every STABLE newer than the baseline plus a one-line alpha-awareness note (it version-compares tags, so skipped numbers are handled; the per-alpha firehose is collapsed to a count):

node .claude/skills/codex-changelog/get-releases.mjs        # reads .last-checked-version
# or with a user-supplied baseline: node .claude/skills/codex-changelog/get-releases.mjs 0.142.5

Read the Bash output directly:

  • LATEST_STABLE=<x.y.z> — the newest stable → this is LATEST_STABLE.
  • ALPHAS=<n> newer[ (latest rust-v…-alpha.NN)] — the alpha-awareness line; surface it verbatim.
  • NOTE: window … — appears only if the 40-release window didn't reach STORED (older stables may be missing; note it rather than assuming full coverage).
  • then, per new stable (newest first): === <tag> (<name>) === followed by its notes.

If gh fails (exit 1), fall back to WebFetch https://github.com/openai/codex/releases and read the same stable notes.

Step 3 — Up-to-date short-circuit

If the output is (up to date — no new stable since <STORED>) (i.e. LATEST_STABLE == STORED, no newer stable), report "already current at STORED" — still surface the ALPHAS= line if newer alphas exist — and STOP. Do not rewrite the state file.

Step 4 — Map the collected stable notes to the bridge

The helper already printed each new stable's notes (Step 2). Classify each relevant item into one of four buckets. Report an item only if it lands on the bridge surface below; never assert impact without checking the anchor file.

Buckets:

  • 🔧 Actionable — needs a change in this repo (bridge argv, sandbox handling, parser).
  • ✅ Favorable — Codex now does natively what the bridge worked around; no action, note it.
  • 🟢 Safe / no-op — plausibly relevant but verified harmless for the bridge.
  • ℹ️ Informational — worth knowing, behavior-neutral.

Bridge surface checklist (what to read to confirm impact):

SurfaceWhat to checkRepo anchors
Spawn argvcodex exec --sandbox read-only - (stdin prompt) shape; flag renames/removals/additionsscripts/codex-bridge.mjs, scripts/codex/args.mjs, scripts/codex/codex.mjs
Sandbox invariant--sandbox read-only (fresh) + -c sandbox_mode=read-only (resume) semantics; whether resume inherits sandboxscripts/codex/*.mjs, CLAUDE.md (Sandbox invariant)
exec / resumecodex exec + codex exec resume behavior; thread/session semantics; the removed native exec review subcommandscripts/codex/resume.mjs, docs/decisions.md
Output parsingstdout JSON/JSONL shape + fields the bridge parses (thread-id, token counts)scripts/codex/codex.mjs (parseCodexJsonl)
Config overrides-c key=value behaviorscripts/codex/args.mjs
Min version / prereqscodex-cli minimum-version requirementCLAUDE.md, scripts/setup-doctor.mjs

For any flagged-but-uncertain item, deepen (read the anchor file / the full release note) before concluding.

Step 5 — Report

Lead with a one-line TL;DR answering "does anything need action?". Then the four buckets (omit empty ones or say "none"), plus the alpha-awareness line. Cite repo files as clickable [path](path) links. Keep it tight — a bridge-maintainer triage, not a release-notes reprint.

Step 6 — Advance the baseline

After reporting, offer to update .last-checked-version to LATEST_STABLE (default: yes). On confirmation, overwrite the file with LATEST_STABLE + newline. Skip the bump if the user wants to keep the current baseline for re-review. Never bump before reporting.

Notes

  • Read-only + report only — never edits the bridge as a side effect. Any actionable finding is handed back to the user to act on (e.g. via hyper-plan).
  • Companion to the cc-changelog skill (sibling under .claude/skills/) — same shape, different source (GitHub Releases via gh vs a raw CHANGELOG.md).

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.