easy-ui-mcp: Skill for Claude Code

Role: Delivery Report Renderer

You hand a task's outcome to a reader as a link, instead of a TASK_GUIDE they'd have to know how to read. This is the companion to thinking-report (Stage 0.5–2 reasoning) and html-report (Stage 4 review scores) — this one covers Stage 5 delivery, and it carries no scored dimension, risk percentage, or findings table. A Delivery Report demonstrates what shipped; it does not assess it (DDR-0003 decision 3).

Activation

Invoked by the Supervisor at Stage 5, after Skill({ skill: "verify" }) passes and before merge.

Skill({ skill: "delivery-report", args: "task=T053 guide=tasks/TASK_GUIDE_T053.md branch=t053-demonstration-block" })

Workflow

1. Run the renderer

The parser and rendering logic live in one place — .claude/skills/delivery-report/render.py — so the same code path serves both guide flavors (implementation and bugfix). T053 gave both flavors identical Demonstration field names and ordering specifically so this renderer never needs a flavor-specific branch. If you find yourself about to special-case a flavor here, stop — that means the two block shapes have drifted apart, and it's a Supervisor decision, not a workaround.

python3 .claude/skills/delivery-report/render.py <TASK_ID> <guide-path> <branch>

This prints the complete rendered HTML to stdout and the save path (reports/delivery-report_<branch>_<timestamp>.html) to stderr. The script already writes the file — you do not need to re-save it.

2. What the renderer does (for context, not to reimplement)

• Resolves both reviewer-filled sections guide first, then the sibling tasks/TASK_REVIEW_Txxx.md (T064), through the one shared resolver .claude/hooks/lib/guide_sections.py. A pre-T064 guide that still carries them inline renders exactly as it always did; a split pair renders identically. A guide whose section is only a > **Moved.** pointer counts as not carrying it, so resolution falls through. Pass the guide path — the renderer looks for the review file beside it.
• Parses the ## Demonstration section for BEFORE/AFTER/DELTA using one regex shared by both flavors.
• If the guide's BEFORE points at the bugfix flavor's "Phase 1 repro loop" by name, resolves it from the Evidence table's Repro loop row rather than printing the pointer text raw.
• Computes the Evidence-table completion count (filled / total, plus N/A count) generically over whatever rows are present — 9 rows on the implementation flavor, 12 on the bugfix flavor.
• WITNESS is never read from the guide as free text. It is always derived from memory/event-trace/<task>.jsonl (AC7) — a count of tool-call records and their timestamp span, not a name. If no trace file or no records exist, it renders as explicitly underived — never a guessed or fabricated name. Do not override this with a typed value.
• If neither the guide nor the review file has a ## Demonstration section at all (a guide predating T053, or a split pair whose review file was never created), the report renders that gap explicitly instead of crashing.
• If BEFORE/AFTER/DELTA is blank in the guide, the report renders that gap explicitly rather than omitting the row or silently succeeding.
• Field text is inserted inside <pre> without manual HTML-escaping, matching the established html-report findings convention.