performance-review

argument-hint: ' ' description: Produce an executive-grade performance review for a single team member. name: performance-review

Performance Review

Produce an executive-grade HTML review of one team member for one review period — delivered in January, May, or September, each covering the preceding four months — grounded in Jira, GitHub, and git evidence.

References

• ../../rules/team.md — roster. Resolve the caller's input to exactly one member; take the GitHub handle, Jira account ID, and role from here.
• ../../rules/jira-rest-api.md — Jira auth (JIRA_API_USER / JIRA_API_TOKEN); never the Atlassian MCP.
• graph.md — the provenance graph every agent reads from and appends to (nodes, edges, id namespaces, the two write primitives).
• rating-calibration.md — role-indexed calibration used by the Group, Axes, and Summary agents.
• querying.md — how an agent pulls extra Jira/GitHub context, with guardrails.

Inputs

Both come from ${ARGUMENTS}. If either is missing, ambiguous, or malformed, ask the caller — do not guess.

• Team member — resolve against the roster to exactly one member → handle, jira_account_id, role, fte, display name.
• Period — yyyy-jan / yyyy-may / yyyy-sep (e.g. 2026-may), resolved to the window and a delivery label (e.g. "May 2026"):
  • yyyy-may → yyyy-01-01 … yyyy-05-01 (Jan–Apr)
  • yyyy-sep → yyyy-05-01 … yyyy-09-01 (May–Aug)
  • yyyy-jan → (yyyy-1)-09-01 … yyyy-01-01 (Sep–Dec of the previous year)
  • If absent, default to the most recent of the three delivery months on or before today.

Orchestration

Five agents build the review on the shared graph (see graph.md). Each is a general-purpose subagent with its own folder under agents/ (an AGENT.md + any scripts); everything for a run lives in output/<handle>/ and the scripts take --dir output/<handle>. The orchestrator only sequences them — it never loads an artifact's contents.

Below, <dir> = output/<handle>.

1. Resolve inputs (above) — the only values the orchestrator holds.
2. Run the agents in order, handing each its AGENT.md + graph.md + <dir> + the values it needs: Collect (agents/0-collect/, also handle/jira_account_id/since/until) → Group (agents/1-group/, role + fte) → Axes (agents/2-axes/, role + fte) → Summary (agents/3-summary/, role + fte) → Render (agents/4-render/, the person details). The grade-producing agents (Group, Axes, Summary) all take fte and apply the capacity-normalization rule in rating-calibration.md (grades scale to capacity; scores stay absolute). Confirm each agent's output exists before starting the next.
3. Verify: ./graph.py --dir <dir> validate passes; report <dir>/review.html and the rating.