SuperAPI AI Directory

wisdom-reflect

Agents & models

View repository

Updated 10-23 23:07

About this skill

Objective

At each configured reflection boundary, route the queued context to the read-only coco-wisdom-reflector agent. Its output is a proposal only. Before any write, validate structural completeness, safety, and autonomous authorship; then route accepted intents through the deterministic promoter.

Required Contract

  • Accept only add, merge, patch, drop-support-file, or delete intents with reason and evidence.
  • Reject targets authored by a human or installed from a marketplace.
  • Append every accepted write to the pattern's .ledger.jsonl with a content-addressed evidence reference.
  • Recall probationary patterns normally, but exclude them from capacity and archival until the tier maturity gate is met.
  • Measure survival by calls over relevant requests. Never use wall-clock age.
  • Archive the lowest-adherence mature patterns when the project or global cap is exceeded. Never delete pattern history.

Exit Criteria

  • The reflector has no write tools.
  • Every mutation passes the promoter gate and creates a ledger entry.
  • Project and global tiers use separate maturity gates and capacity caps.

Anti-Rationalization

Shortcut / Temptation Why It Fails
Let the reflector apply its own proposal Proposer/writer separation is the core safety boundary.
Archive based on age Seasonal patterns must be judged only when opportunities arise.