/agent-shims
Diagnose, inspect, install, update, or remove the complete Codex Rig user-agent shim roster. Use when the user invokes agent-shims with exactly one action—doctor, status, install, or remove—or asks to manage Codex Rig role agents safely.
$ npx -y skills add Borda/AI-Rig --skill agent-shims --agent claude-codeHow it fires
How this skill gets triggered: by you, by Claude, or both.
- Fires itselfAuto-invocation. Claude auto-loads it when your prompt matches the work.
- You can call itInvoke it directly when you want it.
- Slash command
/agent-shims
Context preview
The summary Claude sees to decide when to auto-load this skill.
Diagnose, inspect, install, update, or remove the complete Codex Rig user-agent shim roster. Use when the user invokes agent-shims with exactly one action—doctor, status, install, or remove—or asks to manage Codex Rig role agents safely.
SKILL.md
agent-shims.SKILL.mdname: agent-shims
description: Diagnose, inspect, install, update, or remove the complete Codex Rig user-agent shim roster. Use when the user invokes agent-shims with exactly one action—doctor, status, install, or remove—or asks to manage Codex Rig role agents safely.
Agent Shims
This is an experimental lifecycle tool. Installing authenticated standalone agent TOML does not prove that the active collaboration interface can select that custom profile. A task name or child path matching the role name is not proof. Use installed shims only when the runtime exposes an explicit custom-agent selector and observed child metadata plus verifier output prove selection; otherwise use blank-agent role-card injection.
Accept exactly one action: `doctor`, `status`, `install`, or `remove`. Reject missing, extra, or unknown arguments without writes.
Locate `../../scripts/manage_role_agents.py` relative to this installed `SKILL.md`. Run it with the current Python 3.10+ interpreter and the selected action. Do not copy the manager, resolve it through a source checkout, or edit generated agent files directly.
- `doctor`: run the read-only live prerequisite check and render its JSON result for a person.
- `status`: run the read-only health and installed-roster summary and render its JSON result for a person.
- `install`: report the stable platform block. Do not request approval; no new shim plan or write is allowed.
- `remove`: use the same exact-digest flow to remove every intact managed shim. Never delete by filename prefix or marker alone.
Preserve the manager's exit contract: `0` success/converged, `2` usage, `3` cancelled, `4` drift/conflict, `5` prerequisite blocked, `6` untrusted state, `7` internal/recovery failure. Do not retry a mutating action after codes `4`, `6`, or `7`; report the evidence and keep files untouched.
For `doctor` and `status`, do not return only the raw JSON or a generic safety label:
1. Lead with `Healthy`, `Degraded`, or `Blocked`, followed by the first non-pass check and its exact detail. 2. List every other non-pass check once, then summarize `state`, `targets`, `recovery`, and namespace candidates. 3. State `No files changed.` 4. Give the narrow safe next step. For package or active-package failures, refresh or reinstall Codex Rig and start a fresh session. For executable failures, report the selected path and rerun from a fresh session with stable Python and Codex selection. For permission, owner, type, or link failures, inspect only the named path and verify its metadata before changing anything. For corrupt, inconsistent, modified, or foreign evidence, back it up and do not adopt, edit, or delete it automatically. For recognized recovery residue, use `remove` and review its authenticated approval digest.
Never recommend recursive `chmod`, `chown`, deletion, or link replacement from a diagnostic alone. A protected agent target may be readable by other users, but it must be owned by the current user and have no group/world write or special permission bits. Private lifecycle state remains exact mode `0700`.
After successful install, update, or removal, tell the user to start a fresh Codex session. Thin shims intentionally depend on the installed plugin cache; uninstalling the plugin makes remaining shims unavailable until safely removed or reinstalled.
Read more
name: agent-shims description: Diagnose, inspect, install, update, or remove the complete Codex Rig user-agent shim roster. Use when the user invokes agent-shims with exactly one action—doctor, status, install, or remove—or asks to manage Codex Rig role agents safely.
Agent Shims
This is an experimental lifecycle tool. Installing authenticated standalone agent TOML does not prove that the active collaboration interface can select that custom profile. A task name or child path matching the role name is not proof. Use installed shims only when the runtime exposes an explicit custom-agent selector and observed child metadata plus verifier output prove selection; otherwise use blank-agent role-card injection.
Accept exactly one action: `doctor`, `status`, `install`, or `remove`. Reject missing, extra, or unknown arguments without writes.
Locate `../../scripts/manage_role_agents.py` relative to this installed `SKILL.md`. Run it with the current Python 3.10+ interpreter and the selected action. Do not copy the manager, resolve it through a source checkout, or edit generated agent files directly.
- `doctor`: run the read-only live prerequisite check and render its JSON result for a person.
- `status`: run the read-only health and installed-roster summary and render its JSON result for a person.
- `install`: report the stable platform block. Do not request approval; no new shim plan or write is allowed.
- `remove`: use the same exact-digest flow to remove every intact managed shim. Never delete by filename prefix or marker alone.
Preserve the manager's exit contract: `0` success/converged, `2` usage, `3` cancelled, `4` drift/conflict, `5` prerequisite blocked, `6` untrusted state, `7` internal/recovery failure. Do not retry a mutating action after codes `4`, `6`, or `7`; report the evidence and keep files untouched.
For `doctor` and `status`, do not return only the raw JSON or a generic safety label:
1. Lead with `Healthy`, `Degraded`, or `Blocked`, followed by the first non-pass check and its exact detail. 2. List every other non-pass check once, then summarize `state`, `targets`, `recovery`, and namespace candidates. 3. State `No files changed.` 4. Give the narrow safe next step. For package or active-package failures, refresh or reinstall Codex Rig and start a fresh session. For executable failures, report the selected path and rerun from a fresh session with stable Python and Codex selection. For permission, owner, type, or link failures, inspect only the named path and verify its metadata before changing anything. For corrupt, inconsistent, modified, or foreign evidence, back it up and do not adopt, edit, or delete it automatically. For recognized recovery residue, use `remove` and review its authenticated approval digest.
Never recommend recursive `chmod`, `chown`, deletion, or link replacement from a diagnostic alone. A protected agent target may be readable by other users, but it must be owned by the current user and have no group/world write or special permission bits. Private lifecycle state remains exact mode `0700`.
After successful install, update, or removal, tell the user to start a fresh Codex session. Thin shims intentionally depend on the installed plugin cache; uninstalling the plugin makes remaining shims unavailable until safely removed or reinstalled.
Specialist-agent infrastructure for Python/ML OSS — the scaffolding that lets you maintain at scale without becoming a full-time reviewer.
Repo: Borda/AI-Rig
Other skills on ai-rig.
- /debug
Investigation-first debugging — gather evidence, form confirmed root-cause hypothesis, hand off to fix mode with diagnosis file. TRIGGER when: user reports a symptom or failing test with Python traceback, or asks to investigate a runtime/CI failure with reproducible evidence;
Open skill - /feature
TDD-first feature development — crystallise API as a demo test, drive implementation to pass it, run quality stack and progressive review loop. TRIGGER when: user asks to build new functionality, add a capability, or implement a feature in a Python project; phrases: \"add X\",
Open skill - /fix
Reproduce-first bug resolution — capture bug in failing regression test, apply minimal fix, run quality stack and review loop. TRIGGER when: user reports a bug, regression, or unexpected behaviour in Python code with a traceback, failing test, or issue number; phrases: \"fix
Open skill - /plan
Analysis-only planning — classify and scope a task without writing code; outputs a structured plan to .plans/active/. TRIGGER when: user wants to understand scope and risks before implementation; phrases: \"plan this\", \"scope out X\", \"what would it take to Y\", \"analyse
Open skill - /refactor
Test-first refactoring — audit coverage, add characterization tests, apply changes with safety net, run quality stack and review loop. TRIGGER when: user wants to restructure existing Python code without changing behaviour; phrases: \"refactor X\", \"clean up Y\", \"extract Z\",
Open skill - /review
Multi-agent code review of local Python files, directories, or the current git diff covering architecture, tests, performance, docs, lint, security, and API design. Scope: Python source files in local working tree. Python-file-free targets (pure JS/TS/Go/Rust projects) are out
Open skill

