update-skill

To automate skill updates from JSONL user feedback logs, parse the accumulated entries to identify recurring issues, generate targeted SKILL.md edit proposals, and archive the processed feedback. This pipeline summarizes evaluation distributions and creates versioned changelog entries for local skill repositories.

Generating changelog entries from user feedback involves analyzing JSONL files to summarize complaints and strengths, then proposing precise skill file modifications with version bumps. This automated process ensures updates are governed by user approval and auditable through the generated changelog.

Yes, this automated skill maintenance process applies to local Claude skill repositories. It requires reading and parsing JSONL feedback files stored locally, allowing maintainers to review post-run evaluation entries and approve or reject suggested edits to the target skill files.

Proposing precise file edits to a SKILL.md based on recurring feedback patterns requires parsing JSONL logs to detect evaluation distributions. The pipeline then summarizes identified issues and strengths into concrete update proposals for the skill files, including description modifications and version bumps.

Yes, the skill update pipeline requires manual approval before modifying skill files. It operates under governed deployment, requesting user approval for each suggested change, archiving processed feedback, and proposing wrapper skills to ensure built-in protections for the local repository.

The limitation of using JSONL files for skill update recommendations is that feedback must be accumulated and stored in this specific JSONL format. The pipeline depends entirely on reading and parsing these local JSONL files to identify patterns and propose file edits, requiring structured input.