/etc: search beyond the list

/etc is a marker for open-ended lists. Write X, Y, /etc to tell an agent that X and Y are examples, then ask it to check the broader category for material omissions.

Why it exists

Examples are strong instructions. A trailing "etc." is weak by comparison, so an agent may complete the named items and stop. /etc makes the user's intent plain: check the examples, infer the category, and keep looking.

The skill guards against two common misses:

• Checklist satisfaction: the agent treats the named items as the entire scope.
• Near-neighbor generalization: the agent finds more of the same kind but misses other kinds of problems.

The idea is motivated by work on instruction-induction conflict. Camassa and Shiller found that explicitly identifying the status of instructions improved adherence in their experiments. Their paper studies instruction status, not this marker specifically: "Do as I Say, Not as I Do: Instruction-Induction Conflict in LLMs" (2026).

What /etc means

Given check timeouts, retries, /etc, the agent should:

1. Check timeouts and retries.
2. Infer the broader category, in this case reliability.
3. Look outside the examples for backoff, circuit breakers, silent failures, load behavior, setting interactions, and other relevant omissions.
4. Say what it checked beyond the original list.

The marker expands what the agent considers. It does not grant permission for unrelated changes, destructive actions, external communication, or much higher cost. An explicit exhaustive instruction still wins.

Install

The repository is both a Claude Code marketplace and a Codex marketplace. Each installs the same skill from plugins/etc-plugin/skills/etc.

Claude Code

/plugin marketplace add henryzd8/etc-skill
/plugin install etc-plugin@etc-skill

The installed skill is namespaced as /etc-plugin:etc. It can also trigger from a prompt whose open-ended list ends in /etc.

Codex

codex plugin marketplace add henryzd8/etc-skill
codex plugin add etc-plugin@etc-skill

Invoke it explicitly with $etc, or end an open-ended list with /etc for implicit invocation.

Manual skill install

If the host supports Agent Skills but not plugin marketplaces, copy the skill folder into its skills directory. For example:

# Codex, user level
cp -r plugins/etc-plugin/skills/etc ~/.codex/skills/etc

# Claude Code, user level
cp -r plugins/etc-plugin/skills/etc ~/.claude/skills/etc

Repository layout

<repo root>/
├── .agents/plugins/marketplace.json          # Codex marketplace
├── .claude-plugin/marketplace.json           # Claude marketplace
├── plugins/etc-plugin/
│   ├── .claude-plugin/plugin.json            # Claude plugin metadata
│   ├── .codex-plugin/plugin.json             # Codex plugin metadata
│   └── skills/etc/
│       ├── SKILL.md                          # portable skill instructions
│       ├── agents/openai.yaml                 # Codex skill UI metadata
│       ├── evals/evals.json                   # tracked behavioral eval definitions
│       ├── evals/trigger-evals.json           # tracked trigger cases
│       ├── evals/README.md                    # tracked eval instructions
│       └── scripts/check_skill.py             # tracked deterministic checker
├── etc-workspace/                             # local generated eval outputs (gitignored)
├── LICENSE                                   # MIT
└── README.md

The skill folder and its name: etc frontmatter match, as required by the Agent Skills specification. Eval definitions, the checker, and their README are repository files intended to be reviewed and shipped. Runs and reports written under etc-workspace/ are local working data and are deliberately gitignored; they are not part of the plugin package.

Usage notes

• Only the explicit /etc marker triggers this skill. Plain etc., ..., and Unix paths such as /etc/hosts do not.
• Use it for open-ended lists. If the list is exhaustive, say so.
• Weak examples still steer an agent. Remove a bad example instead of expecting the marker to cancel it out.
• The skill's metadata is available for discovery. The full instructions enter context when the skill is selected.

Releasing

Before publishing:

python3 plugins/etc-plugin/skills/etc/scripts/check_skill.py
gh skill publish --dry-run
python3 -m json.tool .claude-plugin/marketplace.json
python3 -m json.tool .agents/plugins/marketplace.json
python3 -m json.tool plugins/etc-plugin/.claude-plugin/plugin.json
python3 -m json.tool plugins/etc-plugin/.codex-plugin/plugin.json
git diff --check

The checker is also the local test command; run it from the repository root or pass the script's full path from elsewhere. gh skill publish --dry-run validates publication without submitting it. The marketplace files point at this repository, so a repository release also includes a normal commit and push of the tracked plugin, manifest, documentation, and eval files; generated run artifacts are not a build product. Confirm that generated output remains local with git check-ignore -v etc-workspace/ before a release. Bump the version in both plugin manifests for a user-visible release.

The skill also includes paired behavioral cases and trigger near-misses under plugins/etc-plugin/skills/etc/evals. Run marker and baseline prompts in clean, separate agent sessions with the same model. Use multiple repetitions before treating a coverage difference as stable.

License

MIT. See LICENSE.