Grill

SkillDev tools

Clarify research intent and produce draft readiness for human exit.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the Grill skill

What this skill tells your AI

The instructions your AI receives, as published by linzhe001/harness-research in .agents/skills/grill/SKILL.md and read by ahel’s review.

Use $grill for: Clarify research intent and produce draft readiness for human exit.

Runtime Contract

  1. Read harness_core/recipes/grill.yaml and identify the current Action.
  2. Resolve Capability Packs from that Action plus the active Assurance Profile.
  3. Load only the resolved Packs. Ordinary Actions allow 3; declared high-risk Actions allow 5 and must record the escalation reason.
  4. Execute, validate postconditions, and record semantic Runtime Events and Gate Evidence. Do not infer success from artifact existence.
  5. After durable changes under docs/, run python tooling/harness.py docs sync --json; this establishes or validates the live docs link. Git hooks only repair the link after checkout, merge, commit, or rewrite.
  6. Pause only at a declared interrupt or human-stop boundary.

Actions:

  • grill.clarify: research-framing
  • grill.readiness: research-framing, policy-and-approval
  • grill.handoff: research-framing, policy-and-approval

Grill Handoff Invariants

Datasets need a direct downloadable source; baselines need a code repository URL. Put non-executable items and baseline_repo_missing in the Baseline Source Ledger instead of guessing. The Execution Intent Ledger uses kind: policy with hf_access_policy, non_hf_registration_policy, baseline_clone_policy, and baseline_clone_scope. Draft outputs are docs/05_intake/Research_Intent_Draft.md, docs/05_intake/Grill_Round_Log.md, and docs/05_intake/Execution_Readiness_Packet.md. After accepted handoff, run init-project update-from-grill to refresh AGENTS.md, CLAUDE.md, and README.md while preserving ## Custom.

Default profile: reproducible.

Start interactively with python tooling/harness.py grill --json; use --auto for a bounded fresh Agent per Action. Inspect with python tooling/harness.py status --json. Legacy internal sources are inactive migration/debug references, not normal routes.

Signals

GitHub stars
86
Forks
9
Last commit
Aug 2026
Advanced
Catalog kind
skill
Gateway key
grill-linzhe001
Source
github.com/linzhe001/harness-research