Skip to content

Latest commit

 

History

History
24 lines (17 loc) · 3.16 KB

File metadata and controls

24 lines (17 loc) · 3.16 KB

Harness ↔ OpenGrimoire audit alignment (Part B)

Purpose: One row per Part B agent-native dimension from the OpenGrimoire audit (OpenHarness bundle), mapping what we improve or maintain here to concrete paths—without an API. Keeps backlog and audit language aligned when you work in this repo.

Normative external audit (example): OpenGrimoire docs/audit/ (e.g. agent_native_opengrimoire_2026-03-24.md), section Part B — OpenHarness.

# Part B principle What we improve / maintain in OpenHarness Primary paths
1 Action parity Same scripts documented for humans and agents; index stays current CHEATSHEET.md (Agent invocation index), capabilities.harness.yaml, scripts/, AGENT_NATIVE_CHECKLIST.md
2 Tools as primitives Skills as composable primitives (SKILL.md) .cursor/skills/, .cursor/skills/README.md
3 Context injection Handoff and state shape; flow documentation HANDOFF_FLOW.md, state/, portable .cursor/state/ (see state/README.md)
4 Shared workspace Files on disk for human + agent; boundary docs PUBLIC_AND_PRIVATE_HARNESS.md, repo root layout
5 CRUD completeness N/A (no entity API)—honest; state is file CRUD via editor state/README.md
6 UI integration N/A (no product UI in harness)
7 Capability discovery Entry points: README, skills index, checklist, YAML manifest README.md, .cursor/skills/README.md, AGENT_NATIVE_CHECKLIST.md, capabilities.harness.yaml
8 Prompt-native Continue prompt + handoff as prompt-forward workflows HANDOFF_FLOW.md, state/continue_prompt.txt, scripts/copy_continue_prompt.*

Portfolio (MiscRepos): .cursor/docs/AGENT_NATIVE_CHECKLIST.md is an entry stub only (links to this repo’s normative AGENT_NATIVE_CHECKLIST.md + AGENT_NATIVE_CHECKLIST_MISCOPS.md). Do not treat the stub as the full checklist body.

Related: Structured manifest capabilities.harness.yaml; python scripts/verify_script_index.py (YAML ↔ disk ↔ CHEATSHEET backticks); python scripts/verify_skills_readme.py (README vs SKILL.md descriptions).

Backlog (agent-native audit follow-up)

Thin MCP over harness scripts (parity, no new business logic): Optional future work — an MCP server that lists and runs only allowlisted scripts from capabilities.harness.yaml / CHEATSHEET, subprocess from repo root, no arbitrary shell. Would complement terminal parity for agents that only expose MCP. Not implemented here; tracked in BACKLOG.md. Security and allowlisting would be documented alongside MCP_PRIVATE_HOST.md.