v0.3.0 — Card-format reports

v0.3.0 — Card-format reports

Single biggest change to skill-reviewer since v0.1. Reports go from an 8-section structure to a 4-section card structure. Each finding now appears exactly once. Word budgets shrank ~50% across all lenses.

What changed in the output

	v0.2 (before)	v0.3 (now)
Top-level sections	8 (TL;DR / Per-file / Cross-file / Quality signals / Recommended actions / Verification / Files referenced / Second-opinion)	4 (TL;DR / Strengths / Findings / Patterns) + Next step + conditional Second-opinion
Times each finding appeared	3-5 (per-file + cross-file + quality signals + actions table + verification)	1 (the card)
Per-file Summary paragraphs	Required	Removed
Cross-file taxonomy	8 fixed subsections	Free-form Patterns, 3-5 bullets max
Recommended Actions table	Required	Removed (cards ARE the table, prose-shaped)
Verification section	Required, 100-200 words	Removed (per-fix verification lives in the card's Fix line)
Severity legend position	Inside TL;DR	Single-line footer after Next step
Strengths section position	Position 5 of 8	Position 2 (right after TL;DR)
Word budget (plugin shape)	~3500 target / 5000 cap	~1500 target / 2500 cap

Why

Under v0.2, each finding appeared 3-5 times across overlapping sections. Authors reported reconciling five views of the same item to figure out what to fix. Cards collapse that: one finding, one block, every detail a reader needs to act.

The Recommended Actions table from v0.2 was already the most-used section — readers skipped straight to it. v0.3 promotes that pattern to the primary findings vehicle.

What's preserved

• Severity colors (🔴/🟡/🟢/⚪), effort buckets (Small/Medium/Large), quick-win tags
• File:line citations on every claim
• 🟠 for strengths (positive indicator, not severity)
• The "Generalizes:" clause on strengths (high-value, kept verbatim)
• All injection-handling and second-opinion banners
• Every existing lens variant (full / discoverability / safety / architecture / parseability / tests / quick)
• All subcommands (review / summary / lenses / detect / --version)

What you need to do

If you're already using skill-reviewer: nothing. Run your normal /skill-reviewer review <path> invocations; the output structure will be different. No CLI changes. No subcommand changes.

If you re-installed since v0.2.1, run /plugin marketplace remove skill-reviewer + /plugin marketplace add Terryc21/skill-reviewer + /plugin install skill-reviewer@skill-reviewer to refresh the marketplace cache. (The cache doesn't update on /plugin marketplace update; that's a Claude Code quirk, not a skill-reviewer issue.)

Documentation

• New: reference/output-format.md § What changed in v0.3 (preserves the rationale)
• New: reference/output-format.md § Migration from v0.2 (one-time, for skill-reviewer authors) — will be removed in v0.4
• Updated: examples/sample-report-unforget.md is now in v0.3 card format. Compare with git history (git show v0.2.0:skills/skill-reviewer/examples/sample-report-unforget.md) to see the v0.2 shape side-by-side.
• Updated: docs/DESIGN.md has a new "Release history" section.

Compatibility

• Backwards-compatible at the input/CLI surface. All subcommands, lens flags, second-opinion, and detect work identically.
• Breaking change in output shape. If you have downstream tooling that parsed v0.2's Recommended Actions table by column position, it will need updating. Cards use a different shape (severity in heading, file:line + effort + quick-win in citation line).
• Old reports stay valid. Reports produced under v0.2 remain readable. Nothing erases.

Acknowledgment

Format change emerged from a live conversation where the author requested simpler finding reports while reviewing the bug-echo skill. The conversation went: "cognitive load is too high" → "what if findings were one card each?" → draft proposed format → spec rewrite → this release. The bug-echo audit that triggered the conversation is at https://github.com/Terryc21/bug-echo/releases/tag/v1.1.1 (which is itself a card-format-shaped commit history, predating v0.3).