Loading source
Pulling the file list, source metadata, and syntax-aware rendering for this listing.
Source from repo
A comprehensive collection of Agent Skills for context engineering, multi-agent architectures, and production agent systems.
Files
Skill
Size
Entrypoint
Format
Open file
Syntax-highlighted preview of this file as included in the skill package.
researcher/mechanisms/README.md
1# Mechanism Registry23The mechanism registry is the durable index of accepted patterns that can update skills. It exists so novelty checks compare proposed behavior changes, not broad keyword overlap.45Each line in `registry.jsonl` is one accepted mechanism with:67- `mechanism_id`: stable kebab-case identifier8- `owning_skill`: published skill that owns the pattern9- `activation_scenario`: situation where the mechanism applies10- `behavior_change`: what a future agent should do differently11- `evidence`: source URLs or repo artifacts supporting the mechanism12- `failure_modes`: failures the mechanism prevents13- `status`: `accepted`, `candidate`, `deprecated`, or `rejected`1415Only `accepted` and `candidate` mechanisms participate in novelty checks. Rejected mechanisms should remain in run logs or rejected proposal files unless they are useful enough to prevent repeated rediscovery.1617## Promotion Flow1819Runs propose mechanisms in `proposals/mechanism-proposal.jsonl`. Promotion is gated:20211. The run must pass run-readiness validation for `accepted` or `candidate` mechanisms.222. A human reviewer must be recorded.233. Accepted and candidate mechanisms append to `registry.jsonl`.244. Every promotion appends an event to `ledgers/accepted.jsonl`.255. Rejected mechanisms append to `ledgers/rejected.jsonl` so future agents do not rediscover them.26