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.
In this repository22 other files it gives its agents.
- ground-specSkill
- initiative-feasibilitySkill
- learning-extractSkill
- loop-specSkill
- pre-implementSkill
- purge-initiative-artifacts-appSkill
- spec-draftSkill
- spec-implementation-planSkill
- spec-technical-reviewSkill
- ensure-repo-graphSkill
- post-product-questionsSkill
- prd-codebase-mapSkill
- review-product-questionsSkill
- commit-workspaceSkill
- create-board-ticketsSkill
- open-draft-prSkill
- prd-impact-mapSkill
- prd-thinkSkill
- purge-initiative-artifacts-metaSkill
- review-findingsSkill
- update-documentsSkill
- validate-requirementsSkill
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
- Very specific: line 30
If the user names only an INIT, look under reports_dir for {INIT}-prd-think*.md.
- Distinctive: line 40
Do not read one candidate in order to fill gaps in the other.
- Distinctive: line 53
≥2 files: No comparison without both (or all) paths.
- Very specific: line 57
Evidence: Every bar is PASS / FAIL / N/A plus id + quote.
- Very specific: line 58
If you would guess to fill a spec row, that bar is FAIL.
- Very specific: line 59
Same rubric, same weights: Do not add ad-hoc criteria mid-run.
- Very specific: line 60
Do not prefer the later file, the longer file, or the one named prd-think.
- Very specific: line 62
/prd-criticis secondary: If installed, run it per file after the delivery-bar pass. - Distinctive: line 92
Do not mention other candidates in this pass.
- Very specific: line 96
If /prd-critic is installed, run it on each file.
- Very specific: line 131
Never treat Ci-wins as permission to validate.
- Very specific: line 148
When not orchestrated, skip baton write unless handoff_path is bound.
≈ “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.
Files
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)
- Two or more PRD paths — e.g.
{INIT}-prd-think.mdand{INIT}-prd-think-2.md, or a think candidate vsprd/{INIT}.md. If the user names only an INIT, look underreports_dirfor{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.” - Initiative id — from filenames or ask.
- 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). - Layout —
reports_dirfrom.harness/profile.yaml(meta defaultprd/reports/).
Do not read one candidate in order to fill gaps in the other.
Outputs
| File | Role |
{reports_dir}/{INIT}-prd-quality.md | Independent scores + comparison |
Overwrite that one report path. Never write *-revN. Never edit the
input PRDs or prd/{INIT}.md.
NON-NEGOTIABLE
- ≥2 files. No comparison without both (or all) paths.
- Blind first. Score each file in isolation against references/delivery-bar.md before any winner language.
- Evidence. Every bar is
PASS/FAIL/N/Aplus id + quote. If you would guess to fill a spec row, that bar isFAIL. - 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. /prd-criticis 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.- Do not invoke
/prd,/prd-think,/prd-development,/prd-impact-map,/validate-requirements,/commit-workspace,/open-draft-pr. - Do not promote.
Companion
| Skill | Role | If missing |
/prd-critic | Prose / clarity / metrics second opinion | Skip |
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:
- Fill B1–B15.
- Count material FAILs.
- List spec-lane guesses (rows
/spec-draftcould 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.
| Handover | When |
yes | That file has zero material FAILs (ready-to-validate) |
no | Any 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):
CibeatsCjonly 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. ACi-winsfile with material FAILs is stillhandover: 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:
- Persist
{reports_dir}/{INIT}-prd-quality.md. - Overwrite
handoff_pathwith thehandoff:envelope when bound. Leaving the baton empty is a failed stage for automated consumers. - Derive
next_candidatesandhuman_checkpointfrom pinned rootworkflow.yamlfor(stage: prd-quality, outcome)perprayog-skills/references/handoff-envelope.md(Derive from pinned workflow). Sethuman_checkpoint: trueonly when the resolved next node'stypeishuman-checkpoint.
When not orchestrated, skip baton write unless handoff_path is bound.
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
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.

