What problem does it solve? Repeated skill failures often indicate the skill itself needs a patch, but letting an agent modify its own skills directly is unsafe. This Skill turns Memory skill telemetry into concrete, reviewable improvement proposals so fixes are applied only after explicit human approval. ## Core Features & Use Cases - Telemetry-driven proposals: Detects frequently-failing skills from partitioned skill telemetry rollups and drafts proposals with failure evidence, fingerprints, and priority scores. - Approval-gated patching: Applies a structured json memory-skill-patch block only when --yes is present and the oldString matches exactly once in the target file. - Proposal lifecycle management: Lists, shows, applies, and archives pending proposals under .red/memory/proposals/ with deterministic fingerprint deduplication. - Use Case: A skill keeps failing at the verification stage across sessions. Run the proposal surface to generate a proposal with recent failure evidence and a draft patch targeting the Verification section, review it, then apply it with an explicit confirmation flag. ## Quick Start Ask the agent to generate skill improvement proposals from memory telemetry in JSON mode, then review and apply the approved proposal with the confirmation flag.
