thin-pointer

Worked Example: Thin-Pointer Multi-Runtime Instructions

One canonical instruction tree; each agent harness gets a short pointer instead of a forked copy of the same workflow prose. Drift between AGENTS.md, Cursor rules, Codex skills, and other entry files is a silent governance failure.

Comparative exemplars (ideas only): santifer/career-ops, MadsLorentzen/ai-job-search. This fixture is minimal documentation — not a claim that every private app already dual-runs two harnesses.

Path	Role
canonical/review-checklist.md	Full SoT for a tiny review workflow
wrappers/AGENTS.snippet.md	Short AGENTS-style pointer
wrappers/cursor-rule.snippet.mdc	Short Cursor-rule pointer

python3 -m pytest tests/test_thin_pointer_example.py -q

Rule of thumb

Put here	Put in the canonical file
Path to the SoT + one sentence	Numbered steps, rubrics, long policy
Harness-specific discovery metadata (globs, frontmatter)	Shared behavioral content

Trace

Artifact	Location
Rule	AGENTS.md — Thin-pointer multi-runtime instructions
Skills vs standards layers	docs/agent-skills-integration.md
ROADMAP	v0.9