/report

/report — Structured Summary of Recent Work

You are generating a structured report for the user about recent multi-step work in this conversation.

Step 1 — Load the forge-report skill

Read the skill router: plugins/dev-toolkit/skills/forge-report/SKILL.md

If the skill is not found at that path (e.g. user installed dev-toolkit globally), it should be auto-loaded from the installation. Don't fabricate format — always pull from the skill.

Step 2 — Pick the right template

Scan the conversation for the dominant task type. Use this decision tree:

1. Created/built something new (PR, files, feature, plugin) → build-summary
2. Reviewed/audited existing code (security, architecture, quality) → audit-summary
3. Made a product/architecture decision (chose between alternatives, recorded ADR) → decision-summary
4. Debugged or fixed an incident (broken pipeline, prod issue, mysterious bug) → incident-summary
5. Refactored or migrated (framework upgrade, restructure, schema change) → migration-summary

If multiple apply, pick the one whose next step is most actionable for the user.

If none apply (just Q&A, explanation, single edit) — say so:

"Recent work doesn't meet structured-report criteria. Here's a brief summary instead: ..."

Step 3 — Read the template + required sections

For the chosen template:

• sections/01-templates/<template>.md — type-specific structure
• sections/02-required-sections/ — TL;DR, not-done, reversibility, drift-risks, next-steps (all mandatory)
• sections/00-anchors/ — status icons, TL;DR rules, confidence labels

Step 4 — Generate the report

Apply the template strictly:

• TL;DR at the very top, 1-3 lines, ≤80 chars per line
• Use single-legend status icons (✅⏳⚠️❌🔵⚪➡️) consistently
• All 5 required sections present (use "—" if literally nothing applies)
• Confidence labels (✅High/🟡Medium/⚠️Assumed) on uncertain claims
• Cycle metadata at bottom (tool calls, files, time)

Step 5 — Self-check against anti-patterns

Before sending, scan the report against:

• sections/03-anti-patterns/wall-of-text.md — break prose into structure
• sections/03-anti-patterns/over-reporting.md — if task was tiny, downscale to inline summary
• sections/03-anti-patterns/duplicate-info.md — collapse repeated facts

Step 6 — Output

Send the report. Don't add "Here's your report:" preamble — the structure is self-evident.

Notes

• This command is explicit — user invoked /report, so always generate (don't skip even if task was small; user asked).
• For automatic generation (without explicit /report), the global CLAUDE.md rule decides — see ~/.claude/CLAUDE.md "Structured Reports" section.
• If the conversation has no multi-step work to report on, say so honestly — do not fabricate content.