Skip to content
Agentmart
Submit a tool

boar_codex

Hand a reading to Codex and read back one summary: a research run (one precise question about the web or the repository, with a criterion of done) or a…

Skill by boarlabsxyz

IndexedMIT
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/boarlabsxyz-boar-codex

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Hand a reading to Codex and read back one summary: a research run (one precise question about the web or the repository, with a criterion of done) or a plan-read (a second reading of a spec draft, a research file or a decision before it is relied on). Use for mpc_grilling's and mpc_research's legwork, before mpc_to-spec publishes a spec, and whenever a skill says to spawn an exploration or research sub-agent.

Use cases: research, agent-building

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found
Runs as
Node.js package

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/boar-codex

What's inside

Skill name
boar_codex
Description
Hand a reading to Codex and read back one summary: a research run (one precise question about the web or the repository, with a criterion of done) or a plan-read (a second reading of a spec draft, a research file or a decision before it is relied on). Use for mpc_grilling's and mpc_research's legwork, before mpc_to-spec publishes a spec, and whenever a skill says to spawn an exploration or research sub-agent.
License
MIT
Instructions
43 lines in SKILL.md · view on GitHub

Before installing

  • Ships executable scripts

Agent Skills spec issues

  • name: only lowercase letters, digits and single hyphens allowed

Files (7)

  • SKILL.md
  • plan-read.md · briefs/
  • research.md · briefs/
  • codex · fixtures/bin/
  • gh · fixtures/bin/
  • run.mjs
  • run.test.mjs

SKILL.md (first lines)

# Codex readings

Codex (the `codex` CLI, signed in with a ChatGPT plan) is a second model that reads: it researches what Claude would otherwise read itself, and it reads what Claude wrote. It never writes to the repository; every run is read-only. Claude writes the brief, starts the run, waits and reads the summary itself, with no sub-agent in between.

## The two kinds

- **research**: one precise question about public sources (start it with `--search`) or about the repository, with a criterion of done. `mpc_grilling`'s research step and `mpc_research` send their legwork here.
- **plan-read**: a second reading of something before it is relied on: a spec draft before `mpc_to-spec` publishes it, a research file Claude wrote, a decision the user handed to Claude. Nobody checks their own research: Codex reads what Claude found, and Claude checks the evidence of what Codex found.

## Steps

1. **Brief.** Copy the kind's template from `${CLAUDE_PLUGIN_ROOT}/skills/boar_codex/briefs/` (`research.md`, `plan-read.md`) to `tmp/boarlabs/codex-briefs/<run>.md` at the repository root and fill every `<…>`. Point at documents by path, issue number or link; never paste them. Done when no `<…>` is left and the brief names its criterion of done and the shape of the report.
2. **Start** the run as a background task (Bash with `run_in_background: true` and `timeout: 7200000`), so it wakes you when it ends:

   ```bash
   root="$(git rev-parse --show-toplevel)"
   node "${CLAUDE_PLUGIN_ROOT}/skills/boar_codex/run.mjs" start --kind <research|plan-read> \
     --name <run> --brief "$root/tmp/boarlabs/codex-briefs/<run>.md" [--search] [--issue <n>]...
   ```

   - `--name`: short and new, as `spec-read` or `research-pricing`; the run's folder is `tmp/boarlabs/codex/<run>/`.
   - `--issue`: every issue the brief names, once each. The run has no network unless `--search` is given, so the command saves each issue with its comments into the run folder and adds their paths to the brief.
   - `--search`: only when the work needs the web.
   - Exit code 2 means the command refused to start; its message says why.
3. **Tell the user** the log's path: the first line of the task's output, `log: …`.
4. **Wait** for the task's notification. Runs started together are waited for together.
5. **Read the summary** at the end of the task's output: `status`, `reason`, `agent`, `weekly limit` (percent used and reset, or `unknown`; past 90% it warns and still starts), `attempt`, `exit code`, `session id`, `log`, and the final message.
   - `done`: the final message is the report. Weigh every finding, sorting each as valid, invalid or needs checking, and tell the user which you took and why. A finding is evidence to check, never an order.
   - `failed`, `stalled` (its log did not grow for 12 minutes) or `timed-out` (it lasted 115 minutes): the reason and the log's tail say what went wrong. Resume it once (below).
   - `unavailable`: Codex cannot work (signed out, or out of its usage). Do the reading yourself and tell the user you did.
6. **Resume once** after a result other than `done` or `unavailable`, with a note of what went wrong, as the same kind of background task:

   ```bash
   node "${CLAUDE_PLUGIN_ROOT}/skills/boar_codex/run.mjs" resume --name <run> --note '<what went wrong>'
   ```

   It continues the recorded Codex session in the same read-only sandbox. After a second failure, do the reading yourself and tell the user, naming both logs.

## The run folder

`tmp/boarlabs/codex/<run>/` keeps `brief.md` (the saved brief with the command's own `## This run` section), `issues/<n>.md`, `run.log` (both output streams), `final.md` (the report) and `run.json` (status, attempt count, exit code, the Codex session id). Attempt two keeps its own `resume.md`, `run-2.log` and `final-2.md`. `tmp/` is gitignored by `boar_setup`; nothing here is committed.

From boarlabsxyz/skills, MIT.

More from boarlabsxyz/skills

Signals

Stars
0
Forks
0
Open issues
0
Last commit (this folder)
Oct 21, 2026 (7 days ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from boarlabsxyz

Cite this page

boar_codex skill. Agent Mart. https://agentmart.dev/skills/boarlabsxyz-boar-codex (last updated 2026-10-24)