sjawhar-legion-126: skill(test.md) — evidence gate, handoff enforcement

[sjawhar]

Closes #126

Summary

Strengthens the test workflow in two areas:

1. Evidence Gate (Change 7) — Elevates evidence requirements from guidance to a hard gate in Section 5. Every acceptance criterion must now include a concrete artifact (screenshot, command output, log, or code-ref). Local file paths are explicitly prohibited. PR template placeholders updated from [output/screenshot] to [REQUIRED: screenshot/output/log/code-ref]. New Common Mistakes entry added.

2. Handoff Enforcement (Change 8) — Adds MUST-attempt language for handoff writes in Section 6.5, acknowledging the partially broken plumbing (Planner handoff data doesn't reach tester or reviewer directly #124). Pre-signal checklist added to Section 7 requiring PR results posted and handoff write attempted before adding labels.

Both .opencode/ and .claude/ copies are identical (hard-linked files, same inode).

[sjawhar]

Behavioral Test Results

Summary

PASS — 12/12 acceptance criteria verified

Results

Change 7: Evidence Gate Strengthening

Criterion	Status	Evidence
Bold evidence gate callout in Section 5	✅	grep -n 'EVIDENCE GATE' → line 169, within Section 5 (lines 154–182). Blockquote format with bold label.
PR template marks evidence as REQUIRED	✅	grep -c 'REQUIRED: screenshot/output/log/code-ref' → 2 (GitHub template line 197, Linear template line 243)
Allows "Verified by code inspection: [file:line]"	✅	Present in EVIDENCE GATE callout: "Verified by code inspection: [file:line]" at line 169
Allows one artifact for multiple related criteria	✅	Present in EVIDENCE GATE callout: "One artifact may cover multiple related criteria if explicitly noted." at line 169
New Common Mistakes entry	✅	grep -n 'Posting results without evidence' → line 332, last row of Common Mistakes table
Explicitly prohibits local file paths as evidence	✅	grep -c 'Local file paths' → 2 (EVIDENCE GATE callout line 169 + Common Mistakes entry line 332)

Change 8: Test Handoff Enforcement

Criterion	Status	Evidence
Section 6.5 uses "MUST attempt" language	✅	grep -n 'MUST.*attempt' → line 229 in Section 6.5 (starts line 213): You **MUST** attempt the handoff write
Section 7 has pre-signal checklist (steps 6 & 6.5)	✅	grep -A5 'Pre-signal checklist' → line 265 in Section 7 (starts line 259), references "step 6" and "step 6.5"
|| true preserved	✅	grep -n '|| true' → line 218 (code block) and line 229 (explanation text). Not removed or altered.
Distinction clear: attempt mandatory, success non-blocking	✅	Line 229: "MUST attempt" + "
Acknowledges handoff plumbing not fully operational	✅	grep -n '#124' → line 229: "handoff plumbing between phases may not be fully operational yet (#124)"

Dual Directory Mirroring

Criterion	Status	Evidence
Both directories identical	✅	diff returns no output. ls -li confirms same inode (90987533) — files are hard-linked.

Test Quality Critique

N/A — markdown-only changes, no code or tests to critique.

Documentation Feedback

• The EVIDENCE GATE callout reads clearly and is unambiguous about what qualifies as evidence
• The pre-signal checklist is concise and actionable
• The Planner handoff data doesn't reach tester or reviewer directly #124 reference helpfully sets expectations about handoff plumbing state

Observations

• Plan deviation (minor): Plan specified **MANDATORY ATTEMPT:** as a bold label prefix (matching the **EVIDENCE GATE:** pattern); implementer used inline bold **MUST** instead. The acceptance criteria specify "MUST attempt language" not the specific label, so this passes. A consistent label style (**MANDATORY ATTEMPT:**) would improve scannability.
• Plan deviation (minor): Plan included expanded "What does NOT count as evidence" bullet section (3 bullet points); implementer inlined the prohibition into the blockquote and Common Mistakes entry. More concise, still meets all acceptance criteria.
• Plan deviation (minor): Pre-signal checklist simplified from plan (no ✅ checkmarks, no explanatory parentheticals). Still references steps 6 and 6.5 as required.
• Hard-link approach: Both .opencode/ and .claude/ share inode 90987533. Only .opencode/ appears in the PR diff, which is correct — editing one file updates both. This is sound but worth noting for reviewers unfamiliar with the setup.

[sjawhar]

Review Summary

CRITICAL (P1): 0 issues
IMPORTANT (P2): 0 issues
MINOR (P3): 3 suggestions

Verdict: Approved to merge — All 11 acceptance criteria met, CI green (lint/test/typecheck), dual directory mirroring verified (hard-linked inode 90987533).

---

Acceptance Criteria Verification

Change 7: Evidence Gate Strengthening

Criterion	Status
Bold evidence gate callout in Section 5	✅ Blockquote with bold label, correctly placed after evidence types list
PR template marks evidence as REQUIRED	✅ Both GitHub and Linear templates updated
Allows "Verified by code inspection: [file:line]"	✅ Present in EVIDENCE GATE text
Allows one artifact for multiple related criteria	✅ Present in EVIDENCE GATE text
New Common Mistakes entry	✅ Last row of Common Mistakes table
Explicitly prohibits local file paths	✅ In EVIDENCE GATE + Common Mistakes entry

Change 8: Test Handoff Enforcement

Criterion	Status
Section 6.5 uses "MUST attempt" language	✅ You **MUST** attempt the handoff write
Section 7 has pre-signal checklist (steps 6 & 6.5)	✅ Both steps referenced
`
Distinction clear: attempt mandatory, success non-blocking	✅ Clear language separating obligation from error handling
Acknowledges handoff plumbing not fully operational	✅ References #124

Dual Directory Mirroring

Criterion	Status
Both directories identical	✅ Hard-linked (same inode), verified with diff

CI Status

All checks passing: lint ✅ | test ✅ | typecheck ✅

Minor Suggestions (P3)

See inline comments.

[sjawhar]

No blocking issues. All acceptance criteria met. Three minor style suggestions inline.

[sjawhar]

💡 Suggestion (P3)

The plan specified a separate "What does NOT count as evidence" section with bullet points and an example path (/tmp/screenshot.png). The inline approach is more concise, but the expanded format with concrete anti-examples is more scannable for testers who need to quickly self-check. Not blocking — the acceptance criteria are met either way.

[sjawhar]

💡 Suggestion (P3)

Style inconsistency: Section 5 uses a labeled blockquote callout (> **EVIDENCE GATE:**) but this section uses plain paragraph with inline bold (You **MUST** attempt). The plan specified **MANDATORY ATTEMPT:** as a labeled prefix to match the EVIDENCE GATE pattern. Using a consistent callout style (e.g., > **MANDATORY ATTEMPT:** You MUST attempt...) would improve document scannability. Not blocking.

[sjawhar]

😹 Nitpick (P3)

Double blank line before the pre-signal checklist (lines 262-263). Single blank line is the standard markdown separator.