No dates, no version commitments. Items ship when they are well-scoped and tested. Contributions are welcome — see CONTRIBUTING.md.
There is currently no way to see which craftsman skills, agents, or hooks actually fired in a given session. docs/results.md is explicit that no corpus of real before/after sessions exists yet — the plugin is new and external field examples haven't accumulated.
Three approaches that could address this, each with real tradeoffs:
(a) Opt-in session transcript scanning. A local script that parses Claude's conversation export and surfaces which craftsman tools were invoked. Advantage: post-hoc, no runtime friction. Tradeoff: conversation exports may not be available in all Claude configurations, and parsing them reliably requires knowing Claude's exact export format, which can change.
(b) Append-on-fire hook log. A hook that writes a timestamped line to a local log file
whenever it fires — e.g., ~/.claude/craftsman-memory/hook-log.md. Advantage: lightweight,
always-on audit trail with no self-reporting involved. Tradeoff: only captures hook events, not
skill or agent invocations, and adds a write side effect to every hook run.
(c) /craftsman:session-summary command. A command that prompts Claude to reflect at the end
of a session on which craftsman disciplines were applied. Advantage: covers agents and skills
that leave no filesystem trace. Tradeoff: relies on Claude's self-report of its own behavior,
which is not the same as an objective log — the accuracy of that reflection is unverified.
This item needs a scoping conversation before implementation. The tradeoffs — privacy implications of a persistent log, friction cost of an end-of-session prompt, and the correctness risk of self-reported telemetry — must be settled before any code is written. Open an issue or Discussion to start that conversation.