self-improve
/self-improve
Self-Improvement (SIP) turns each skill run into a chance for that skill to get better. When a participating skill finishes, SIP checks whether the run surfaced a concrete weakness in the skill — and if so, drafts a specific improvement and surfaces it for the handler to approve. It is the per-skill, in-the-moment, signal-triggered counterpart to the deliberate batch reflection of /insight-skill-gap and /insight-retro.
SIP is the structural mirror of NSP: NSP runs after a skill and asks "what should you do next?"; SIP runs after a skill and asks "how should this skill be better next time?"
SIP is invoked two ways: automatically, when a skill carrying the SIP opt-in block completes; or directly, as /self-improve [skill] to review a specific skill against the most recent run.
Structured execution spec:
protocol.yon. Read it for the canonical rules and step sequence; this file is explanation. The two must stay in sync — if you edit one, update the other and refresh the@STAMPdate.
Conservative by default. A missed minor improvement is cheaper than nag fatigue. SIP surfaces nothing unless a signal is concrete, material (it cost something this run — a deviation, a guess, a halt, or a wrong output) or recurring within the session (the same signal seen ≥2× this session), and a specific fix is ready. Silence is the common case.
Signal taxonomy — what counts as a real signal
SIP fires only on concrete signals tied to this run. Vibes ("this could be cleaner") do not qualify. The qualifying signals: