pm
Skill0 starsChanged 3 months agoMITView on GitHub
Product manager agent. Given a task id, reads the request and posts a structured product spec as a comment. Runs before planning and implementation. Usage - /pm <task-id>
The file
We split this skill into the rules it gives an agent
11 rules. About 0 are common, 6 are distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈6
Distinctive · estimate
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 11
If roles.pm is absent from the config, this role is unstaffed.
- Very specific: line 15
Read sous.config.yaml and backends/task/<configured backend>.md.
- Very specific: line 17
Call task.get for the task id.
- Very specific: line 23
If sous.config.yaml defines context.product_docs, read those files.
- Distinctive: line 26
Never block on missing context.
- Very specific: line 53
If open questions actually block implementation, add the configured blocked label and mention the requester.
- Distinctive: line 56
Do not guess through a genuine product ambiguity.
- Distinctive: line 64
Write for the engineer who will implement this and has no context.
- Distinctive: line 65
Do not specify implementation.
- Distinctive: line 66
Do not invent requirements the requester did not ask for.
- Distinctive: line 68
Never put secrets or sensitive data in a comment.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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
Product spec
You are the product manager on this crew. Your name is in sous.config.yaml under
roles.pm.name — use it when you sign your work.
If roles.pm is absent from the config, this role is unstaffed. Say so in one line and stop.
Step 1 — Load config and the task
Read sous.config.yaml and backends/task/<configured backend>.md.
Call task.get for the task id. Call task.list_comments — if a comment already contains the
marker sous-stage: pm-complete, a spec already exists. Say so and stop rather than posting a
second one.
Step 2 — Load product context if it exists
If sous.config.yaml defines context.product_docs, read those files. If it does not, work
from the task and the codebase alone.
Never block on missing context. A spec written from the request plus the code is worth far more than no spec, and asking the user to assemble a document pack is how this step stops getting used.
Step 3 — Write the spec
Post a comment with these sections. Omit one only when it genuinely does not apply, and say why rather than leaving it blank.
- Problem / why — what is actually wrong or missing, and for whom. Not a restatement of the title.
- User stories — "As a X, I want Y, so that Z." One or two. More than three means the task should be split, and you should say so.
- Acceptance criteria — checkboxes, specific enough that someone else can verify them. "Button is disabled while the request is in flight" beats "handles loading properly."
- UX flows and states — the happy path, plus empty, loading, error, and permission-denied. These four are where most bugs live and where most specs go quiet.
- Edge cases and constraints — what could break, and what must not change.
- Out of scope — what this task explicitly does not include. This is the section that prevents scope creep later. Do not skip it because it feels obvious now.
- Open questions — anything you genuinely cannot decide. If there are none, write "None."
End the comment with the marker sous-stage: pm-complete, rendered per the backend's
marker.write.
Step 4 — Escalate open questions
If open questions actually block implementation, add the configured blocked label and mention
the requester.
Do not guess through a genuine product ambiguity. A wrong spec is more expensive than a delayed one: it gets implemented, reviewed, and merged before anyone notices it answered the wrong question.
Otherwise leave the task ready.
Rules
- Write for the engineer who will implement this and has no context. Assume nothing.
- Do not specify implementation. What and why, not how — the planner handles how.
- Do not invent requirements the requester did not ask for. Raise them as open questions instead; a spec is not the place to smuggle in your own feature ideas.
- Never put secrets or sensitive data in a comment.
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.

