audit-accomplishments

Audit Accomplishments

Mine a person's contribution evidence across every source where their work lives, over a configurable window, and emit cited achievement summaries for review prep. Collection only — this skill does not draft the review.

Sibling of audit-history: that skill mines sessions to improve agent config; this one mines work across external systems to surface achievements. The local-history + memory discovery is shared — reuse audit-history Phase 1 rather than re-deriving it.

Inputs

• window — default last 6 months. Accept an override as dates (2026-01-01..2026-06-24) or duration (3 months). Compute concrete start/end dates up front and state them.
• sources — default all (below). Accept a subset override.
• taxonomy (optional) — a review template's dimensions / career-level areas, supplied as an arg or a path. When absent, group by discovered theme only; do not invent a rubric.
• output dir — default ./review-material/ in the cwd.

Guardrails

• Read-only across every external system. Never post, edit, comment, transition, or mutate.
• Cite everything. Each claim carries a re-fetchable reference (URL, file:line, ticket id, transcript UUID) so it can be expanded during drafting.
• Completion over creation. Weight what was finished in the window (PR merged, issue resolved), not what was opened or merely discussed.
• No invention or embellishment. If a section has no evidence, say so. Stay grounded in what was found.

Phase 1 — Discover identities & sources (mechanical, read-only)

Resolve who the person is on each source and where to look. Do not summarize yet.

• GitHub — gh auth status enumerates logged-in hosts/accounts. Record each host + login. The person may have multiple (e.g. work + personal); mine all unless overridden.
• Jira — call the Atlassian MCP atlassianUserInfo (server name varies by install) for the account id; getAccessibleAtlassianResources for the cloud id(s).
• Notion — get the self user via the Notion MCP (get-users / self lookup).
• Slack — resolve the logged-in user_id (the search tool reports it). Mine two axes, since neither alone is complete: (a) authored — from:<@USERID> after:<start> to enumerate the channels the person posts in (a global from:me misses DMs and needs real keywords, not stopwords), then drive Phase 2 per-channel; (b) mentions of the person — <@USERID> after:<start> -from:<@USERID> and to:<@USERID>, surfacing where others defer to, route work to, or @-mention them. Prefer the <@USERID> token over literal-name search — a common first name is noisy and misses @-mentions entirely.
• Local agent histories + memory — reuse audit-history Phase 1: Claude Code transcripts under ~/.claude/projects/*, Cursor under ~/.cursor/projects/*, memory under ~/.claude/projects/*/memory/. Filter to files modified in the window.