issue-to-prd
Use when converting a new GitHub issue into a structured PRD, or posting clarifying questions when the issue is underspecified.
How do I install this agent skill?
npx skills add https://github.com/fellowship-dev/dogfooded-skills --skill issue-to-prdIs this agent skill safe to install?
- Gen Agent Trust Hubpass
The skill issue-to-prd automates the conversion of GitHub issues into structured Product Requirement Documents (PRDs) through a multi-stage workflow. It includes safety guards to avoid processing sensitive issues, performs context audits, and can dispatch remote worker missions to build UI prototypes. The primary security considerations involve the ingestion of untrusted data from GitHub issues, which is subsequently used to generate and publish content, and the use of internal vendor APIs for impact measurement. Command execution and network access are restricted to vendor-controlled tools and infrastructure.
- Socketwarn
1 alert: gptAnomaly
- Snykwarn
Risk: MEDIUM · 1 issue
What does this agent skill do?
issue-to-prd
ICM procedure: convert a new GitHub issue into a structured PRD or post clarifying questions.
Invocation
/issue-to-prd {repo} {issue_number}
Example: /issue-to-prd fellowship-dev/pylot 123
When to use
- A new issue arrives and needs structuring (triggered by
challenge-new-issueevent rule) - A human answered
open-questionsand re-challenge fires (triggered byrechallenge-after-answers) - Manual invocation to structure a specific issue
What it does
11-stage ICM procedure:
- 00-automation-guard — refuse
no-automation/epic/ closed issues, before reading anything - 01-read-issue — fetch issue data (title, body, labels, comments)
- 01b-triage-challenge — search for an existing mechanism serving the same need and check the
ask against its smallest version →
delete-retire/close/re-scope/prd; non-prdstops the pipeline (stages 02-07 skipped) - 02-context-completeness — audit implicit context gaps a fresh agent would miss
- 03-assess-clarity — check against 8-section gap checklist →
clearorneeds-questions - 04-failure-modes — predict how an agent can go astray → guardrails for PRD
- 05-test-plan — test strategy + prerequisites (pre-merge and post-merge)
- 05b-prototype-gate — is this UI exploration? →
skip/ask/dispatch/picked - 05c-outcomes-baseline — detect a causal goal/eval/outcome contract in the issue; only then
fetch
GET /outcomes/summaryfor that named metric's baseline (fail-open) - 06-ask-or-structure — decision: post questions (exit) OR draft PRD (Measurable Impact included only when a contract exists, else "not applicable")
- 07-publish — rewrite issue body with PRD, apply labels
Exit paths
- Guard path (stage 00): no reads, no writes,
status=success, stops - Triage path (stage 01b): posts one comment naming the verdict and its cited evidence, applies
wontfixorneeds-rescope, stops (stages 02-07 skipped) — never fires without a citation - Questions path (stage 06): posts GH comment, adds
open-questionslabel, stops - PRD path (stage 07): rewrites issue body, adds
prd-ready(+ready-to-workunless prototype variants are pending)
Prototype variants
Stage 05b can dispatch a mission that pushes proto/<issue>-a|b|c branches and posts one options
comment with screenshots and boot instructions, for the owner to pick from. It is gated hard:
auto-detection can only ever ask a question; only an explicit owner signal (the
prototype-options label, or a "build the prototype" answer) dispatches. See
stages/05b-prototype-gate/CONTEXT.md.
Stage handoffs
Each stage writes to its output/handoff.md. Downstream stages read upstream handoffs.
All output directories at stages/{stage}/output/.
Execution
Run stages sequentially. Read the CONTEXT.md for each stage before executing it.
Stage 00 is a hard gate — if it aborts, run nothing else.
Stage 01b is a hard gate — on any verdict but prd, run nothing after it (stages 02-07 skipped).
Stage 05c always runs (fail-open) — it never gates stage 06.
After stage 06, skip stage 07 if questions were posted.
Reference files
shared/prd-template.md— PRD structure (used by stage 06);## Measurable Impactsection is conditional on stage 05c's contract verdictshared/failure-modes.md— common agent pitfall catalog (used by stage 04)shared/prototype-mission-brief.md— self-contained brief for the variant mission (stage 05b)stages/03-assess-clarity/references/gap-checklist.md— 8-section checkliststages/05b-prototype-gate/references/gate-checklist.md— prototype gate + worked classificationsstages/05c-outcomes-baseline/CONTEXT.md— causal contract detection + scoped, fail-open baseline formatting
How can the creator link this skill?
Add the canonical catalog link to the repository README so users can inspect current installs and available audits. The publishing guide covers the complete discovery path.
<a href="https://skillzs.dev/skills/fellowship-dev/dogfooded-skills/issue-to-prd">View issue-to-prd on skillZs</a>