What problem does it solve? Writing an end-of-session report from memory produces vague, unverifiable claims. This Skill builds the report entirely from measured figures — git commits, merged PRs, GitHub Actions job-minutes, sub-agent token counts, AWS deploy counts, and operator message classifications — so every number in the report is traceable to a command run in the session. ## Core Features & Use Cases - Deterministic report naming: Writes REPORT_SESSION_<id>_<date>.md at the repo root, where the id is derived from the SHA-256 of the session id, so re-running overwrites the same file. - Measured data gathering: Runs git log, gh run list, gh api, and CloudFormation queries to collect commits, PRs, billable job-minutes, deploy counts, and ci stack lifetimes before writing anything. - Structured report format: Produces a result line, a first-person method paragraph, a what-worked table, a losses table with causes, placement against cited external anchors, and ranked improvement suggestions with values. - Use Case: After a long Claude Code session that merged multiple PRs and deployed to production, invoke the skill to produce a costed account of what landed, what it cost in tokens and CI minutes, and which process improvements would remove the measured losses. ## Quick Start Ask the agent to write the session report for this session using the session-report skill.
