prd-to-spec

PRD to SPEC Generator

User-facing entry point for the PRD → SPEC pipeline. The user runs

/prd-to-spec and this skill dispatches to the companion babysitter

process via /babysitter:call (interactive, with breakpoints) or

/babysitter:yolo (non-interactive, auto-approve).

Usage

/prd-to-spec <PRD-path-or-inline-text>

Examples:

• /prd-to-spec docs/active/feature-X-PRD.md
• /prd-to-spec "Build a notification dashboard with real-time alerts"

Companion process

• Library path: library/specializations/product-management/prd-to-spec.js
• Locally installed: ~/.a5c/processes/prd-to-spec.js

The process file is the source of truth for the pipeline. This skill

intentionally stays thin so users only need to remember /prd-to-spec.

What this skill does when invoked

1. Parse input — resolve the PRD path or capture the inline text.
2. Infer featureBranch — read the current git branch via

git branch --show-current. If it is main / develop / dev,

ask the user for a feature-branch name.

1. Detect optional inputs from the project context:

• contextDoc — project conventions doc (default CLAUDE.md)
• archiveDir — past SPEC archive directory (default empty)
• failureLogPath — project-local failure-log.md (default empty)
• secondaryReviewer — hint like codex, gemini, peer (default empty)

1. Ask the user to pick the run mode:

> Run mode for the babysitter prd-to-spec process:

>

> 1. Interactive (/babysitter:call) — recommended. Pauses at

> every breakpoint (discovery review, SPEC approval, etc.) for

> explicit user confirmation.

> 2. Auto / yolo (/babysitter:yolo) — non-interactive. Skips

> all breakpoints; auto-approves every gate. Use only when the

> pipeline is well-tested and the inputs are trusted.

Default to option 1 if the user does not respond.

1. Dispatch by invoking the chosen babysitter command via the Skill tool:

For interactive:

   Skill('babysitter:call', 'Run the product-management/prd-to-spec process with inputs: prdPath=<resolved>, featureBranch=<resolved>, contextDoc=<resolved>, archiveDir=<resolved>, failureLogPath=<resolved>, secondaryReviewer=<resolved>')

For auto / yolo:

   Skill('babysitter:yolo', '<same instruction as above>')

1. Return the SPEC path and the execution prompt produced by the

process to the user.

Fallback (no babysitter runtime)

If neither /babysitter:call nor /babysitter:yolo is available:

1. Read the process source at the location above.
2. Execute each task's agent prompt manually, in phase order.
3. Pause at every ctx.breakpoint(...) for explicit user approval

(interactive equivalent) or skip every breakpoint (yolo equivalent),

matching the mode the user picked in step 4.

1. Return the same outputs the process would have returned.

Optional Integration Hooks

All flow through the process inputs (none required):

• Tracker integration — trackerHint input. Examples: jira

(/jira or acli jira workitem update), linear (CLI / MCP),

gh-issues (gh issue edit).

• Secondary reviewer — secondaryReviewer input. Examples: codex

(/codex:review), gemini (/gemini-review), deep-verify-plan

(/deep-verify-plan), peer (notify a teammate).

• Pipeline framework for Phase 3 verification — examples: Dagster,

Airflow, dbt, Prefect, K8s preview.

• UI framework for Phase 3 verification — examples: Vite, Next,

Storybook.

The process works end-to-end without any of them.

Failure Log Mechanism

The process reads a project-local failure-log.md in Phase 1 to

internalize lessons from prior failed runs. Setup:

• Cross-project lessons: ~/.claude/skills/<your-namespace>/failure-log.md
• Project-specific lessons: <repo-root>/.claude/failure-log.md

Each entry follows:

### [YYYY-MM-DD] <short-pattern-name>

**Context:** <what we tried to do>
**What went wrong:** <the failure mode in 1-2 sentences>
**Constraint going forward:** <the binding rule for future runs>

If the file doesn't exist, the process notes "no prior failures recorded"

and continues.

Pre-Push Personal Docs Check

The process executes this gate as part of Phase 7 (Deliver) of the

generated SPEC. Procedure:

1. List candidates: git diff --name-only origin/<base-branch>...HEAD
2. Classify each file: legitimate / personal-doc / ambiguous
3. Block by default for: SPEC.md, PRD.md, HANDOFF*.md,

SUMMARY.md, -NOTES.md, anything under docs/plans/, ai_docs/,

.claude/scratch/, tmp/, investigation-, analysis-,

playbook-*, AI-generated plans/specs not explicitly approved.

1. Block-and-ask gate: per-file decision of unstage / keep /

gitignore. Do NOT skip even if rushed.

1. Proceed with push only after all blocked files are resolved.

Rules

• The skill itself does not produce the SPEC. It only sets up inputs

and dispatches to the process via /babysitter:call or /babysitter:yolo.

• Default to /babysitter:call (interactive) when in doubt.
• Never bypass user-approval breakpoints in interactive mode — they

are first-class in the process.

• The SPEC must be proportional to the task (the process honors this).