agentleFS
drivestream-labdrivestream-lab/prayog-skills/skills/requirements/prd-quality/SKILL.md

prd-quality

Skill2 starsChanged 4 months agoMITView on GitHub

Score two or more PRD files independently against the spec-lane delivery bar, then compare. Use after /prd-think reruns, or whenever the user passes candidate paths (prd-think series or prd/{INIT}.md). Scientific: blind per-file PASS/FAIL with quoted evidence, then a comparison table. Optional companion /prd-critic is secondary and cannot override a material FAIL. Writes reports_dir/{INIT}-prd-quality.md. Never edits the PRDs. Never run /prd, /prd-think, /prd-impact-map, or forge.

The file

We split this skill into the rules it gives an agent

12 rules. About 0 are common, 3 are distinctive, and 9 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 9

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    If the user names only an INIT, look under reports_dir for {INIT}-prd-think*.md.

    line 30
  2. Distinctive:

    Do not read one candidate in order to fill gaps in the other.

    line 40
  3. Distinctive:

    ≥2 files: No comparison without both (or all) paths.

    line 53
  4. Very specific:

    Evidence: Every bar is PASS / FAIL / N/A plus id + quote.

    line 57
  5. Very specific:

    If you would guess to fill a spec row, that bar is FAIL.

    line 58
  6. Very specific:

    Same rubric, same weights: Do not add ad-hoc criteria mid-run.

    line 59
  7. Very specific:

    Do not prefer the later file, the longer file, or the one named prd-think.

    line 60
  8. Very specific:

    /prd-critic is secondary: If installed, run it per file after the delivery-bar pass.

    line 62
  9. Distinctive:

    Do not mention other candidates in this pass.

    line 92
  10. Very specific:

    If /prd-critic is installed, run it on each file.

    line 96
  11. Very specific:

    Never treat Ci-wins as permission to validate.

    line 131
  12. Very specific:

    When not orchestrated, skip baton write unless handoff_path is bound.

    line 148

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

PRD quality

Score files the user passes, not a mode inside /prd-think.

A good PRD here means /spec-draft can implement it without guessing. Page count, tone, and “looks like a PRD” are not scores.

Pin contracts (read, do not fork):

  • Ids: prayog-skills/references/id-conventions.md
  • Paths: prayog-skills/references/artifact-write-contract.md

Inputs (required)

  1. Two or more PRD paths — e.g. {INIT}-prd-think.md and {INIT}-prd-think-2.md, or a think candidate vs prd/{INIT}.md. If the user names only an INIT, look under reports_dir for {INIT}-prd-think*.md. If fewer than two files exist, ask — do not invent a second document, do not score a singleton “vs imaginary A.”
  2. Initiative id — from filenames or ask.
  3. Brief — (OPTIONAL) outline used to author the candidates. Needed only for bar B12 (whole-product vs brief-shaped). If absent, score B12 as N/A — brief not supplied (not a free PASS).
  4. Layout — reports_dir from .harness/profile.yaml (meta default prd/reports/).

Do not read one candidate in order to fill gaps in the other.

Outputs

FileRole
{reports_dir}/{INIT}-prd-quality.mdIndependent scores + comparison

Overwrite that one report path. Never write *-revN. Never edit the input PRDs or prd/{INIT}.md.

NON-NEGOTIABLE

  1. ≥2 files. No comparison without both (or all) paths.
  2. Blind first. Score each file in isolation against references/delivery-bar.md before any winner language.
  3. Evidence. Every bar is PASS / FAIL / N/A plus id + quote. If you would guess to fill a spec row, that bar is FAIL.
  4. Same rubric, same weights. Do not add ad-hoc criteria mid-run. Do not prefer the later file, the longer file, or the one named prd-think.
  5. /prd-critic is secondary. If installed, run it per file after the delivery-bar pass. Its Build Readiness cannot override a material FAIL. If missing, skip — do not inline a fake critic.
  6. Do not invoke /prd, /prd-think, /prd-development, /prd-impact-map, /validate-requirements, /commit-workspace, /open-draft-pr.
  7. Do not promote.

Companion

SkillRoleIf missing
/prd-criticProse / clarity / metrics second opinionSkip

Process

T0 Resolve files

List the paths. Label them C1, C2, … from filenames only. Tell the user the output path.

T1 Independent score (repeat per file)

For each Ci, in order, with only that file + the rubric + optional brief in mind:

  1. Fill B1–B15.
  2. Count material FAILs.
  3. List spec-lane guesses (rows /spec-draft could not fill).

Do not mention other candidates in this pass.

T2 Secondary critic (optional)

If /prd-critic is installed, run it on each file. Record Strengths / Gaps / Build Readiness under a Secondary (prd-critic) heading. Discard critic claims that contradict a delivery-bar FAIL.

T3 Handover, then compare

Write {reports_dir}/{INIT}-prd-quality.md from references/scorecard-template.md.

Handover is the only signal for validate → review → update. It is per file, not the comparison winner.

HandoverWhen
yesThat file has zero material FAILs (ready-to-validate)
noAny material FAIL (think-again — keep looping /prd-think)

A third (or Nth) think file is just another Ci. Score it the same way. yes on two files with different jobs is not a tie-break — the human picks which named file to promote.

Rank (secondary, after handover):

  • Ci beats Cj only if fewer material FAILs on B3, B4, B6, B9, B11, B12 PASS (or both N/A for the same reason), and B2 equal or better.
  • Rank must not override a no. A Ci-wins file with material FAILs is still handover: no.

Present the Handover table in chat first, then the comparison. A third run is C3; do not overwrite earlier candidates.

Verdict language

Header Verdict = comparative rank only (C1-wins / C2-wins / tie). Header Handover = comma list C{n}: yes|no. Never treat Ci-wins as permission to validate. Never B-wins vs a hidden current PRD unless that file was an input.

Workflow handoff

This skill is human-invoked — it is not a node in pinned workflow.yaml. When orchestrator-bound:

  1. Persist {reports_dir}/{INIT}-prd-quality.md.
  2. Overwrite handoff_path with the handoff: envelope when bound. Leaving the baton empty is a failed stage for automated consumers.
  3. Derive next_candidates and human_checkpoint from pinned root workflow.yaml for (stage: prd-quality, outcome) per prayog-skills/references/handoff-envelope.md (Derive from pinned workflow). Set human_checkpoint: true only when the resolved next node's type is human-checkpoint.

When not orchestrated, skip baton write unless handoff_path is bound.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.