check-deliverable

/check-deliverable — standalone qa gate

Run the existing qa-reviewer agent against ONE finished text deliverable on demand, without a full Ann/Vi orchestration. Reuse qa-reviewer wholesale: add no new checking logic, rewrite nothing, write nothing to disk.

Lane — when this skill, not another

• /check-deliverable (this): checks an existing finished file against the qa gate. One file, one qa-reviewer spawn.
• /ann: produces a deliverable through the full pipeline. Use when the goal is the artefact.
• /ane-voice: audits and rewrites prose voice. Use to fix tone, not to run the full gate.
• /code-review: reviews code diffs, not MEL prose deliverables.

Arguments

/check-deliverable <file> [--tier 1|2] [--subgroup colleague|ma-staff|partner-ngo|management|junior-mel|peer-review]

• <file> (required): path to the deliverable.
• --tier (default 1).
• --subgroup (default colleague).

Derive voice positioning from the subgroup:

• management to directive
• junior-mel to collaborative-pedagogical
• all other subgroups to collaborative

Steps

1. Validate the file. Resolve the path. If the extension is not .md, .txt, or .markdown, STOP and return exactly this, then spawn nothing:

   qa-reviewer can only inspect text. For Word, Excel, or PowerPoint deliverables run the full /ann pipeline so the brand and format checks apply.

2. Spawn qa-reviewer via the Agent tool with subagent_type: "qa-reviewer". Pass this prompt, substituting the bracketed values:

   ## Standing instructions
   Audience tier: [Tier 1|Tier 2]; subgroup: [subgroup]; voice positioning: [positioning]
   
   ## Task
   Standalone single-file review. Review the deliverable at this path:
   [absolute file path]
   
   There is NO Ann plan, NO Definition-of-done criteria, and NO P1 wiki block.
   Cold-load the wiki per your session-start fallback (qa-block-schema.md,
   domain-standards.md, calibration.md) and omit the Definition-of-done subsection.
   Read and Grep the file directly to run every mandatory check (tier register,
   em-dash, citation, lens, power-shift, edge-case probe). Populate the qa_block,
   add reconciliation notes, and close with your VERDICT line.
   

3. Reformat the result BLUF-first:

   • Line 1: the verdict (APPROVED / PASS_WITH_GAPS / REJECTED — reason).
   • Then the load-bearing findings in plain prose: hard FAILs first, then gaps, then soft flags.
   • Then the full verbatim qa_block under a ### Full qa_block heading.

   Reformat only. Never override, soften, or reinterpret a qa-reviewer finding.

Boundaries

• One file, one qa-reviewer spawn. No reader-position-reviewer.
• No prose rewriting (that is /ane-voice).
• No writes to the deliverable or anywhere on disk.