fix(implement): count checkbox markers outside code fences only (#4272)

[ntdatt812]

Closes #4272.

templates/commands/implement.md told the agent to count checkbox markers like this:

- Total items: All lines matching `- [ ]` or `- [X]` or `- [x]`

Every line, fenced blocks included. So a checklist that documents the checkbox format — an example fence showing what - [ ] looks like — reported unchecked items that no reviewer can ever tick, and the gate stops on a non-zero unchecked count. Writing down the format blocked implementation.

templates/commands/clarify.md already had the right rule and the right words for it:

lines matching - [ ], - [x], or - [X] … outside of code fences

So this was also two commands disagreeing about what a checklist item is, which is the part worth fixing rather than just the count.

Only templates/commands/implement.md needed the change: grep for the scan instruction across templates/commands/ and presets/*/commands/ returns exactly these two files, and the lean preset's speckit.implement.md does not carry a counting block.

Test

tests/unit/test_checklist_scan_contract.py turns the rule into a contract rather than a one-time edit. It walks every command template, finds each line that defines what counts as a checkbox marker, and asserts that line also excludes code fences. It is parametrized per instruction, so the failure names the file, the line number and the offending text:

FAILED …::test_marker_scans_exclude_code_fences[implement.md-63-     - Total items: All lines matching `- [ ]` …]

Two details that make it a real guard rather than a passing test:

• a companion test_the_contract_is_actually_stated_somewhere fails if the regex ever matches nothing, so deleting or rewording the scan instruction cannot make the suite green by making the parametrization empty;
• it scans presets/*/commands too, so a preset that grows its own counting block is held to the same rule.

Mutation-checked — restoring the old wording in implement.md fails exactly one case, the implement one, and the other three stay green:

1 failed, 3 passed

python -m pytest tests/unit/test_checklist_scan_contract.py -q
4 passed

No behaviour outside the checklist scan changes: the per-checklist total/checked/unchecked reporting and the status table are untouched.

[Copilot]

🟡 Changes recommended

The contract regex misses checked-only scan instructions.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Pull request overview

Aligns implementation checklist scanning with clarify by excluding fenced examples.

Changes:

• Clarifies checklist counting rules.
• Adds a cross-template contract test.

File summaries

File	Description
templates/commands/implement.md	Excludes fenced checkbox examples.
tests/unit/test_checklist_scan_contract.py	Enforces scan wording across templates.

Review details

• Files reviewed: 2/2 changed files
• Comments generated: 1
• Review effort level: Balanced

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[mnriem]

Please address Copilot feedback. If not applicable, please explain why

[ntdatt812]

Closing in favour of #4520, which consolidates this with #4314 and #4330 as asked on #4460. This fix is carried over unchanged, in its own commits.

Copilot's point is addressed there: the scan-instruction pattern only recognised a definition whose first marker is unchecked, so the - [X] checked-items definition was never guarded, and removing its code-fence exclusion left the suite green (measured: 4/4 passed). It now matches either marker, four definitions are guarded instead of three, and that removal fails.

AI disclosure, per CONTRIBUTING: this comment and the changes it describes were written with Claude Code as a coding agent.