prd-think
Skill2 starsChanged 4 months agoMITView on GitHub
Write a PRD by challenging the brief as a hypothesis, not by filling an outline. Grill job, actors, non-goals, and kill assumption; model the product; red-team; write a candidate as spec-draft input. Product ids from prayog-skills id-conventions. Writes reports_dir/{INIT}-prd-think.md (or -2, -3 on rerun). Use when starting or rewriting a PRD or when an outline must not be treated as baseline. Never overwrite prd/{INIT}.md unless the user authorizes promote. Never score candidates — use /prd-quality. Never run /prd, /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-qualitySkill
- purge-initiative-artifacts-metaSkill
- review-findingsSkill
- update-documentsSkill
- validate-requirementsSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
20 rules. About 0 are common, 8 are distinctive, and 12 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈8
Distinctive · estimate
This author's own choices. Worth a look.
12
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 17
Write a PRD that /spec-draft can implement without guessing.
- Distinctive: line 31
If the pin mount is missing, say so and still follow those rules from a harness skill copy if present.
- Very specific: line 66
Never write prd/{INIT}.md from T4.
- Distinctive: line 77
WHAT not HOW: No routes, payloads, token encoding, framework, cache, or module names in requirement text.
- Very specific: line 84
Capability + journey first-class: Every CAP-* lists ≥1 J-{nn} or explicit defer.
- Very specific: line 88
Silent gaps are failures: If spec-draft would guess, emit OQ-* or grill.
- Very specific: line 94
Whole-product: If the candidate TOC equals the brief's TOC, fail the run.
- Very specific: line 97
Do not score: No delivery-bar table, no winner.
- Distinctive: line 108
After a companion, return here.
- Distinctive: line 109
Write the candidate with references/candidate-schema.md.
- Very specific: line 123
Prefer /grilling on the brief + available context.
- Very specific: line 125
Must resolve or park as OQ-*:
- Distinctive: line 134
Hard stop: if the job and the proposed solution disagree, do not draft REQs.
- Very specific: line 139
Prefer /domain-modeling then /opportunity-solution-tree.
- Distinctive: line 141
Before any REQ table:
- Distinctive: line 150
If multiple outcomes could be this INIT, name them and recommend one.
- Very specific: line 154
Prefer /strategy-red-team.
- Distinctive: line 161
Write the path from T0 using references/candidate-schema.md.
- Very specific: line 167
Stop. Recommend /prd-quality with this file and at least one other PRD path (earlier think candidate, or prd/{INIT}.md if the user wants …
- Very specific: line 204
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 think
Write a PRD that /spec-draft can implement without guessing.
The brief (outline, notes, or spoken ask) is a hypothesis. Expanding its headings is a failed run. Architecture, routes, payloads, frameworks, and module names are a failed run — park those for engineering.
This skill authors. It does not score. After one or more candidates
exist, run /prd-quality and pass the files.
Pin contracts (read, do not fork):
- Ids:
prayog-skills/references/id-conventions.md - Paths / overwrite:
prayog-skills/references/artifact-write-contract.md
If the pin mount is missing, say so and still follow those rules from a harness skill copy if present. Do not invent a second id vocabulary.
This skill does not unlock Gate 1. Typical next: /prd-quality on two
candidates, then user authorizes promote of one file to prd/{INIT}.md,
then /validate-requirements.
Inputs (resolve, do not assume an INIT)
- Brief — (REQUIRED) outline file, pasted ask, or conversation. Prefer
prd/INIT-*-outline.mdorprd/*-outline.mdif the user points at one. - Initiative id —
INIT-*. Ask if missing. Do not invent a programme prefix. - Existing PRD — (OPTIONAL)
prd/{INIT}.md. Context only. Never input for T1. Never edit unless promote is authorized. - Product context — (OPTIONAL) whatever exists:
planning/,config/service-catalog*.yaml, siblingprd/INIT-*.md, ADRs. Absence is not a blocker — record as skipped. - Layout —
reports_dirfrom.harness/profile.yaml(meta defaultprd/reports/per artifact-write-contract).prd_rootdefaultprd/.
Outputs
Live PRD filename is prd/{INIT}.md. Mark conductor output with suffix
-prd-think so it cannot be mistaken for KEEP product.
Series (do not overwrite a previous think candidate):
| First free path | Role |
{reports_dir}/{INIT}-prd-think.md | Run 1 |
{reports_dir}/{INIT}-prd-think-2.md | Run 2 |
{reports_dir}/{INIT}-prd-think-N.md | Run N |
Pick the lowest N whose file does not exist. Never clobber an earlier run.
Never write prd/{INIT}.md from T4.
This series is an experiment exception to “one path per concern”: the
candidates are the evidence /prd-quality compares. Do not use *-revN.
NON-NEGOTIABLE
- Brief is not baseline. Challenge it before filling it.
- Do not edit the brief or an existing
prd/{INIT}.mdduring the run. - Do not invoke
/prd,/prd-development,/to-prd,/prd-impact-map,/commit-workspace,/open-draft-pr,/prd-quality,/prd-critic. - WHAT not HOW. No routes, payloads, token encoding, framework, cache, or module names in requirement text.
- Ids. Assign only product ids from
prayog-skills/references/id-conventions.md(CAP-*,J-{nn},REQ-*canonical,CTR-*,OQ-*). Do not mintFR-*,D-*, bareJ1, or any other namespace. Discovery locks are a numbered table of statements, or they stayOQ-*until they become a CAP/REQ/J. - Capability + journey first-class. Every
CAP-*lists ≥1J-{nn}or explicit defer. Evidence containing live cites ≥1CAP-*and/orJ-*— not bare “live”. Capability and journey are separate prove rungs (prayog-skills/references/quality-confidence-ladder.md). - Silent gaps are failures. If spec-draft would guess, emit
OQ-*or grill. Do not invent a REQ; do not omit the gap. - One question at a time in T1 (or one
/grillingfrontier round). Recommend an answer every time. Wait. - Locks are outputs of this grill. Confirm or reopen locks that are already in the brief. Do not copy another INIT's lock table.
- Whole-product. If the candidate TOC equals the brief's TOC, fail the
run. Consider at least one journey, actor, negative path, or adjacent
surface the brief did not list (keep /
OQ-*/ non-goal). - Do not score. No delivery-bar table, no winner. Point at
/prd-quality.
Companion skills (use if installed)
| Phase | Skill | If missing |
| T1 | /grilling | Inline § Challenge |
| T2 | /domain-modeling | Inline § Model |
| T2 | /opportunity-solution-tree | Inline § Model (OST fork) |
| T3 | /strategy-red-team | Inline § Stress |
After a companion, return here. Companions must not write the PRD. Write the candidate with references/candidate-schema.md.
Process
T0 Gather
Inventory: brief, INIT id, existing PRD, catalog, planning/siblings, existing
{INIT}-prd-think*.md files.
Resolve the next free output path. Tell the user that path and that
prd/{INIT}.md will not be touched.
T1 Challenge
Prefer /grilling on the brief + available context.
Must resolve or park as OQ-*:
- Job / done-in-the-world (who hurts — do not assume hats from a prior INIT)
- Why now / if we do nothing
- Solution vs job (is the ask a proxy?)
- Actors who can actually do what we might assign
- Load-bearing non-goals
- Kill assumption:
Fails if ___
Hard stop: if the job and the proposed solution disagree, do not draft REQs. Present the fork and wait.
T2 Model
Prefer /domain-modeling then /opportunity-solution-tree.
Before any REQ table:
- Domain terms (from T1, not from a template)
- Journeys including ones the brief omitted
- Adjacent surfaces: keep / defer / non-goal
- Seams (
CTR-*seeds) after the split is decided — logical operation + meaning + errors, not URLs - NFR map: specify or N/A with reason
If multiple outcomes could be this INIT, name them and recommend one.
T3 Stress
Prefer /strategy-red-team. Cap ~5 load-bearing claims.
Steelmans → Fails if ___ → cheapest test → kill criteria.
Survivors → REQ or assumption. Casualties → OQ-* or non-goal.
T4 Draft candidate
Write the path from T0 using references/candidate-schema.md.
Mint product ids per id-conventions. Do not copy another INIT, an existing PRD, or an earlier think candidate to fill the page.
Stop. Recommend /prd-quality with this file and at least one other PRD
path (earlier think candidate, or prd/{INIT}.md if the user wants that
comparison).
Promote (explicit authorization only)
On user request only: copy the file they name → prd/{INIT}.md, then
recommend /validate-requirements. Do not run impact-map or forge from
this skill.
Inline protocol (companions missing)
Challenge: job → why now → solution vs job → actors → non-goals → kill assumption. One question at a time; recommend an answer.
Model: terms, journeys (incl. omitted), adjacent keep/defer/non-goal, seams, NFR map. OST fork if multiple outcomes.
Stress: ≤5 load-bearing claims; Fails if ___; cheapest test.
Thinking is done when each REQ-* has condition + observable result without
guessing — or you can name the OQ-* that blocks that row.
Workflow handoff
This skill is human-invoked — it is not a node in pinned workflow.yaml.
When orchestrator-bound:
- Persist the candidate at the T0 path under
reports_dir. - 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-think, 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.

