fix(issue-backlog-stats): expect marketplace-form actions in step-4 evals

[thisisnic]

Summary

• issue-backlog-stats step-4-recommend failed deterministically on case-1-high-untriaged and case-2-high-stale: the model returned top_action in the marketplace form (/magpie-issue:triage, /magpie-issue:stale-sweep) while the fixtures expected bare skill names.
• The skill is right. Its rules table says "Skill names here are the marketplace form" and links the marketplace doc, and the step's output-spec.md describes top_action as a slash command string. The sibling pr-management-stats fixtures already expect the same form.
• The bare names are a leftover from docs: make the marketplace plugin install the documented default #1195: its first commit de-stuttered the fixtures, and a later commit in the same PR ("repair invocation forms the de-stutter sweep broke") reinstated the slash form in the stats dashboards. The fixtures were never re-synced.
• Updates the two expected.json files to the slash form. No skill text changes.

Type of change

• Skill change (.claude/skills/<name>/) — eval fixtures only
• Tool / bridge contract (tools/<system>/*.md)
• Python package (tools/*/ with pyproject.toml)
• Groovy reference impl
• Cross-cutting (RFC, AGENTS.md, sandbox, privacy-LLM)
• Documentation (docs/, README.md, CONTRIBUTING.md)
• Project template (projects/_template/)
• CI / dev loop (prek, workflows, validators)
• Other:

Test plan

Before the change, full suite on main at 399ab1bc:

Model	Runs	Result
claude -p --model claude-opus-5-5	10	step-4-recommend/case-1-high-untriaged and case-2-high-stale FAIL on all 10 runs, identical diff each time; the other 13 cases passed every run

After the change, the step-4-recommend suite with the updated fixtures:

Model	Runs	Result
claude -p --model claude-opus-5-5	3	3 of 3 cases passed on every run

The other suites in this directory are untouched by the change; their 13 cases passed on every one of the 10 before-runs above.

• prek run passes on the changed files
• For skill changes: eval suite passes for the affected skill
  (PYTHONPATH=tools/skill-evals/src python3 -m skill_evals.runner --cli "claude -p --model claude-opus-5-5" tools/skill-evals/evals/issue-backlog-stats/step-4-recommend/, 3 consecutive runs)
• For skill behaviour changes: a new or updated eval fixture is included in this PR — the fixtures are the whole change; no behaviour changed

RFC-AI-0004 compliance

No principles touched — eval expectation data only; no new mutation, host access, or LLM hop. No SKILL.md edit, so no token restamp.

Linked issues

Refs #1626, which records the failing run in the suite's Run history table.

---

Generated-by: Claude Code (Fable 5.1)

🤖 Generated with Claude Code