Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →PRD template and authoring rules for Foreman. Synthesizes a PRD from the approved plan and the grilled decisions and writes it as a local file in the Foreman feature directory. Does not interview the user and does not publish to any external issue tracker.
.claude/skills/visionforge-ou-foreman-to-prd/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 97% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 975% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -29% | 0% |
| case-14 | ✗→✓ | ▲ Improved | 62% | 0% |
| case-16 | ✗→✓ | ▲ Improved | 78% | 0% |
(Adapted from mattpocock/skills to-prd — see NOTICE. Removed: live "check with the user" seam confirmation and the GitHub publish + ready-for-agent label step. Output is a local prd.md file, not a tracker post.)
This skill is the PRD template authority. The foreman-grill-docs skill calls it to produce prd.md. Synthesize from the approved plan, the codebase, and the grilled decisions — do NOT interview anyone.
the project's domain glossary (CONTEXT.md) throughout, and respect ADRs in the area you're touching.
seams; use the highest seam possible. If a new seam is needed, propose it at the highest point you can — and if whether that seam is acceptable is a genuine product/architecture call you cannot settle from the code, add it to the ## Open questions for reviewer block rather than asking interactively.
prd.md into the feature directory using the template below. Begin thefile with the open-questions block (see foreman-grill-docs). Do not publish anywhere and do not apply any labels.
<prd-template>
<unresolved product/architecture questions, or "_None — all resolved._">
The problem the user is facing, from the user's perspective.
The solution to the problem, from the user's perspective.
A LONG, numbered list of user stories, each: As an <actor>, I want a <feature>, so that <benefit>. Extremely extensive — cover all aspects of the feature. These stories are the basis for the slicer (foreman-to-issues) and for the e2e flows, so make each one concrete and verifiable.
For each end-to-end flow a user can perform, list the ordered steps and the observable outcome. These flows are what Foreman's e2e phase will turn into automated tests, so be precise about preconditions, steps, and expected results.
Modules built/modified, interfaces changed, technical clarifications, architectural decisions, schema changes, API contracts, specific interactions. No file paths or code snippets (they go stale) — exception: a prototype-derived snippet that encodes a decision more precisely than prose (state machine, reducer, schema, type shape) may be inlined, trimmed to the decision-rich parts.
What makes a good test here (test external behavior, not implementation details); which modules will be tested; prior art for the tests (similar tests already in the codebase); the test/lint/typecheck commands Foreman will run to verify work.
What is explicitly not part of this PRD.
Anything else worth recording.
</prd-template>
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 19,929 | 5,098 | -74% | 1 | 1 | 0% | 3,371 | 928 | -72% | 0 | 0 | — |
case-02 | fail→fail | 3,545 | 3,259 | -8% | 1 | 1 | 0% | 208 | 877 | +322% | 0 | 0 | — |
case-03 | fail→fail | 35,447 | 6,427 | -82% | 1 | 1 | 0% | 6,002 | 1,008 | -83% | 0 | 0 | — |
case-04 | fail→pass | 6,934 | 7,548 | +9% | 1 | 1 | 0% | 1,034 | 2,032 | +97% | 0 | 0 | — |
case-05 | fail→fail | 20,607 | 21,554 | +5% | 1 | 1 | 0% | 3,001 | 3,859 | +29% | 0 | 0 | — |
case-06 | fail→pass | 3,876 | 9,734 | +151% | 1 | 1 | 0% | 212 | 2,279 | +975% | 0 | 0 | — |
case-07 | fail→pass | 28,541 | 17,206 | -40% | 1 | 1 | 0% | 4,743 | 3,360 | -29% | 0 | 0 | — |
case-08 | fail→fail | 5,637 | 3,100 | -45% | 1 | 1 | 0% | 861 | 1,183 | +37% | 0 | 0 | — |
case-09 | pass→fail | 21,312 | 6,772 | -68% | 1 | 1 | 0% | 4,389 | 1,061 | -76% | 0 | 0 | — |
case-10 | pass→pass | 16,823 | 14,689 | -13% | 1 | 1 | 0% | 2,841 | 3,278 | +15% | 0 | 0 | — |
case-11 | fail→fail | 10,854 | 4,449 | -59% | 1 | 1 | 0% | 1,767 | 993 | -44% | 0 | 0 | — |
case-12 | fail→fail | 15,657 | 5,349 | -66% | 1 | 1 | 0% | 2,617 | 1,029 | -61% | 0 | 0 | — |
case-13 | pass→pass | 24,985 | 18,954 | -24% | 1 | 1 | 0% | 4,173 | 4,142 | -1% | 0 | 0 | — |
case-14 | fail→pass | 15,881 | 22,522 | +42% | 1 | 1 | 0% | 2,321 | 3,751 | +62% | 0 | 0 | — |
case-15 | fail→fail | 22,001 | 13,903 | -37% | 1 | 1 | 0% | 3,369 | 2,727 | -19% | 0 | 0 | — |
case-16 | fail→pass | 21,959 | 37,618 | +71% | 1 | 1 | 0% | 3,857 | 6,860 | +78% | 0 | 0 | — |
case-17 | fail→fail | 22,470 | 16,510 | -27% | 1 | 1 | 0% | 3,385 | 2,904 | -14% | 0 | 0 | — |
case-18 | fail→fail | 15,904 | 3,598 | -77% | 1 | 1 | 0% | 2,636 | 993 | -62% | 0 | 0 | — |
case-19 | pass→fail | 20,294 | 5,849 | -71% | 1 | 1 | 0% | 3,283 | 1,004 | -69% | 0 | 0 | — |
case-20 | fail→fail | 22,676 | 18,963 | -16% | 1 | 1 | 0% | 3,547 | 3,602 | +2% | 0 | 0 | — |
case-21 | pass→fail | 17,588 | 7,197 | -59% | 1 | 1 | 0% | 2,998 | 1,079 | -64% | 0 | 0 | — |
case-22 | fail→fail | 13,097 | 5,993 | -54% | 1 | 1 | 0% | 2,177 | 1,013 | -53% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 11 counted toward the lift figure. The other 11 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +9 percentage points is the difference between those two pass rates over the 11 comparable cases. 5 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.