sidekick / skills
AGENTS.md1 starsChanged 4 months agoMITView on GitHub
Use sidekick doctor ENGINE, then inspect native control evidence with sidekick capabilities ENGINE --model MODEL --json before delegating. Start work with sidekick spawn ENGINE NAME --dir PATH -- "PROMPT"; use separate control axes (--model, --provider, --transport, --effort, --permission, --sandbox, --workspace-trust) instead of treating legacy --mode as a universal semantic. Start one background sidekick wait NAME... --all --json for a batch, continue independent work, and harvest with sidekick result NAME --json. Check sidekick status --json and terminal exitCode; status shows active…
In this repository4 other files it gives its agents.
- AGENTS.mdAGENTS.md
- skills/codex/AGENTS.mdAGENTS.md
- sidekickSkill
- sidekickSkill
The file
Files
Sidekick
Use sidekick doctor ENGINE, then inspect native control evidence with sidekick capabilities ENGINE --model MODEL --json before delegating. Start work with sidekick spawn ENGINE NAME --dir PATH -- "PROMPT"; use separate control axes (--model, --provider, --transport, --effort, --permission, --sandbox, --workspace-trust) instead of treating legacy --mode as a universal semantic. Start one background sidekick wait NAME... --all --json for a batch, continue independent work, and harvest with sidekick result NAME --json. Check sidekick status --json and terminal exitCode; status shows active runs plus bounded recent history unless --all is supplied. Use sidekick clean --dry-run before retention deletion, especially when state diagnostics report legacy or unreadable directories; use sidekick migrate --json to preview legacy conversion and require explicit --apply before mutation. Add --quarantine for unsupported state and migrate NAME --restore --apply to recover a quarantined entry. Continue with sidekick send NAME -- "FOLLOW-UP"; send accepts no control override because persisted controls are immutable. Use sidekick adopt for an existing session, sidekick tail for live output, sidekick cancel for a resumable full-tree stop, and sidekick clean for terminal-run retention. Install harness guidance with sidekick skill install HARNESS.
Each new run records requested, applied, and effective observations. An argv flag is not evidence of provider-reported effective state. Unsupported, malformed, model-dependent-unverified, or runtime-version-mismatched explicit controls fail closed with exit 2 before prompt/cwd/discovery/store/worker side effects. Codex effort is value-gated for any model (minimal|low|medium|high|xhigh|max). Hermes has no public per-run effort flag; do not invent config-home hacks. Devin family slug glm-5.2 and variant UID glm-5-2 are distinct; never silently alias them. normal is the current Devin default alias for native auto; do not use --mode high as an effort setting.
For adversarial review, use a read-only reviewer prompt with explicit no-edit/no-commit/no-push constraints and a catalog-probed model id. For Codex high-effort reviews prefer --effort max or xhigh with an explicit model. For Devin GLM-5.2 High use the currently evidenced model id from sidekick doctor devin / the live catalog (family slug glm-5.2 and variant UID glm-5-2 are distinct; never invent an alias). Require sidekick wait, sidekick result --json, exit code, and independent diff/test verification before accepting a review.
Do not concurrently spawn workers with the same --dir. Harness session discovery and workspace state can be directory-scoped, causing lock contention, accidental serialization, or overlapping edits. For real parallelism, give each worker a separate git worktree and pass that worktree as --dir. If separate worktrees are unavailable or tasks must share a working tree, run one worker at a time and reuse its session with sidekick send.
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.

