Skip to content

RizwanZafaris/product-manager-OS / skills/conductor

conductor

Verified

Interview-driven operation of the six-stage product loop. Use when a user says "start" on a new product, "resume" or "where are we" on an existing one, or asks to be walked through a stage rather than handed a template. Asks one question at a time from the stage's question bank, cross-examines weak answers with a capped challenge grammar, lands every accepted answer in STATE.md and the target template, renders the stage gate on evidence, and never advances past a gate silently.

FileKindSize
SKILL.graph.ymlconfig211 B
SKILL.mdskill md10.7 KB
questions/README.mdmarkdown4.0 KB
questions/build.mdmarkdown5.6 KB
questions/define.mdmarkdown12.7 KB
questions/deliver.mdmarkdown10.4 KB
questions/design.mdmarkdown7.6 KB
questions/discover.mdmarkdown8.4 KB
questions/operate.mdmarkdown9.0 KB

questions/README.md

View on GitHub
---
layer: skills
stage: ALL STAGES
gate: 1
feeds: []
method: ""
aliases: ["Question Banks"]
---
# Question Banks: format and rules

One bank per stage, six banks. Each bank is the normative question set the Conductor runs for that stage, in ID order, under the contract in [../../../os/CONDUCTOR.md](../../../os/CONDUCTOR.md). Banks are data for the interviewer, not prose for the user: the Conductor renders each entry into the four-part question anatomy at ask time, drafting the lettered options from the product's own context unless the entry fixes them.

## Entry format

Every question is one entry with exactly these fields:

```
### <STAGE>-<n>: <handle, three words or fewer>

Ask: <the question, one interrogative sentence, asked as written>
Wrong costs: <one line: the downstream damage of a bad answer here>
Evidence class: <minimum class from the ladder below>
Cross-examine when: <the answer pattern that fires a push> Move: <the grammar move to apply>.
Accept when: <the test an answer must pass, checkable in one read>
Lands in: <workspace-relative path plus section> and STATE.md accepted answers
```

Optional fields, used where they earn their place:

- `Options:` fixed lettered options, only when the choice set is stable across products (stakes, weights, persist-pivot-sunset). Everywhere else options are drafted at runtime and must differ in consequence.
- `Applies:` the knowledge card or skill the question operationalizes, linked, because the Conductor names the framework it is applying and why, every time it applies one.
- `Follow-up on strength:` the harder question a strong answer earns immediately.

## Bank file structure

Each bank opens with a header block: the stage, the gate it feeds, the working skills and agents the Conductor hands off to, and the overlays that can attach. Each bank closes with two sections:

1. **Forced pair.** The two highest-stakes questions for the escape hatch in [../../../os/CONDUCTOR.md](../../../os/CONDUCTOR.md). When a user says "advance anyway", these are forced first, in the order listed.
2. **Gate rendering.** A table mapping every checklist line of the stage's gate in [../../../os/STAGE-GATES.md](../../../os/STAGE-GATES.md) to the bank IDs whose accepted answers evidence it. This is the exit-gate test: a gate line with no accepted, landed answer behind it is marked unknown, and an unknown stops the Conductor exactly as a fail does, though the `pmos` runtime does not enforce that: it records a gate approval without reading these lines, so a gate can read approved with lines unmet or unjudged.

## The evidence ladder

Strongest first. Every `Evidence class` field names its minimum from this list.

1. **Observed behavior**: something a user did, with a date and a place it is recorded.
2. **Artifact**: a document, dataset, ticket, or export a reader could open.
3. **Named commitment**: a person with standing said yes in writing.
4. **Interview claim**: a real person said it, cited by source and date.
5. **Team belief**: goes to the assumptions register, never into a template as fact.

## Rules

- IDs are stable. `Next question` in STATE.md points at an ID; renumbering breaks every resumed journey. New questions append, they do not insert.
- The `Ask` sentence is asked as written. The Conductor may add product context around it, never soften it.
- One decision per entry. An Ask line may join two clauses when they resolve the same decision; an entry that forces two separate decisions is two entries.
- `Lands in` paths are workspace-relative per [../../../os/PRODUCT-WORKSPACE.md](../../../os/PRODUCT-WORKSPACE.md): the filled copy, never the blank under `templates/`.
- A bank edit is a repo change: run `python3 lint.py --os`, then `python3 tools/question_banks.py`, which refuses a bank whose gate-rendering table has more or fewer rows than its gate has checklist lines in [../../../os/STAGE-GATES.md](../../../os/STAGE-GATES.md). It counts the rows and does not read them, so check by eye that each row still paraphrases the line in its position.
slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack."Stars: 198kUpdated Claude CodeCodexVerifiedInstall
ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read.Stars: 196kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report.Stars: 196kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt.Stars: 196kUpdated Claude CodeCodexVerifiedInstall