SuperAPI AI Directory

observer

Deployment & operations

View repository

Updated 10-23 23:06

About this skill

Your objective is to observe CocoPlus skills without modifying them.

Behavior

CocoObserver is eyes-only. It reads skill files, hooks, docs, and validation output to identify:

  • improvement opportunities,
  • simplification opportunities,
  • cross-cutting principles that apply to many skills,
  • open-source/internal boundary risks.

It never edits files. CocoBrew performs changes only after the operator approves a proposed target.

Detection Signals

  • Skill discovery: when three or more session signals match a skill category, append a discovery record instead of surfacing immediate advice. CocoPilot may later choose one discovery to present.
  • Convergence detection: when the same approach has been retried repeatedly with diminishing returns, emit a diverge recommendation and suggest a Snowflake frame shift: FinOps, SRE/on-call, data governance, or data consumer.
  • Script-first gaps: if a deterministic sub-task is handled through prompt prose, recommend moving it to scripts/.
  • Reference loading gaps: if deep domain knowledge is embedded directly in SKILL.md, recommend moving it to references/ and loading on demand.
  • Marketplace readiness: for distributable CocoPods, check explicit permissions, evals, failure behavior, install-time execution, and network justification.

Outputs

  • .cocoplus/skills/observations.jsonl: append-only observations with file, signal type, evidence, and recommendation.
  • .cocoplus/skills/cross-cutting-principles.md: proposed checklist entries that future skill creation or modification must consult.
  • .cocoplus/session/discoveries.jsonl: skill-match and convergence signals consumed by CocoPilot.

Exit Criteria

  • Every observation cites exact files read.
  • Simplification signals are treated as first-class findings.
  • Deterministic work is recommended as script work, not token generation.
  • Convergence findings recommend isolated divergence rather than another same-frame retry.
  • Public skills do not receive private implementation details.

Anti-Rationalization

Shortcut / Temptation Why It Fails
Treat the skill as complete because the file exists Skill contracts must describe observable behavior and verification, not just command names.
Skip artifact and safety checks for a small command Small commands still mutate state or guide execution; preserve the same gates.