Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/ayeo-io-intent-interview
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/intent-interview
What's inside
- Skill name
intent-interview- Description
- Drawing `intent.md` out of a conversation. Load before acceptance criteria exist.
- License
- Apache-2.0
- Instructions
- 136 lines in SKILL.md · view on GitHub
Agent Skills spec issues
- files: referenced but missing: templates/intent.md
Files (2)
- SKILL.md
- role-translation.md
SKILL.md (first lines)
# Intent interview There are two ways an issue gets an `intent.md`, and this skill covers both. **Someone arrives with a document.** It is in Notion, or Jira, or a file called something nobody would guess. `compass intent ingest` has read it and left a snapshot at `intent-source.md`. Your job is to turn it into `intent.md`. **Someone arrives with an idea.** No document at all. Same discipline, longer conversation, nothing to reshape. The rule that makes both safe is the same one, and it is worth stating before anything else. ## Nothing is invented **Every statement in `intent.md` traces to the source or to a recorded answer. There is no third origin.** You can move material, split it, merge it under a template heading, and reformat it. You must not change what it means, and you must not add to it. That last part is the one that will tempt you. A source document with no non-goals is a source document that feels unfinished, and you will be able to think of a perfectly sensible non-goal. **Write it and you have put a product decision into the record that no person made** - and it will read exactly like one they did make, because there is nothing on the page to distinguish them. So: where the source is silent, you ask. Where you ask and nobody answers, the document says so, in words. `compass check` holds this mechanically through `validate_intent_origins` - every section with content must name where it came from. The check is not the reason to do it; it is what stops the reason being forgotten. ## Step aside when there is nothing to ask **Read the source first, and count what it already answers.** If it carries the problem, the outcome, the non-goals, the success signals and a first slice, then there is nothing to draw out. Reshape it into the template's sections, record every section as `from: source`, and hand it over. A person who arrives with a complete source document and gets interrogated about it learns that Compass wastes their time. That is a worse outcome than a thin `intent.md`, because they do not come back. **Say what you are doing:** "Your source document covers everything the template asks for, so I have reshaped it and asked nothing." That sentence is the whole hand-off when the source is good. ## When there is something to ask **One question at a time.** A list of six questions gets one answer to the first and silence for the rest. Ask, wait, use the answer, then ask the next. **Offer choices where the choices are real.** "Is the ranking change in scope, or is this only about latency?" is answerable in a second. "What are your non-goals?" is a blank page with a question mark. Only offer options you would be content with - a menu with one plausible item is a leading question. **Draft the section only after its content is agreed.** Writing first and asking for corrections puts the person in the position of editing your words instead of saying theirs, and most people will accept what is written rather than push back. **Ask about the first slice before you close.** "If only part of this shipped first, which part delivers most of the value?" is the question people are most grateful to have been asked and least likely to volunteer. ## Declining is a complete answer **A person can decline every question and must still get an `intent.md`.** Someone in a hurry, or without the answers to hand, needs to start now and come back later. A loop that blocks them is worse than no interview. When a question is declined: - record it in `interview` with `answer: null` - the question was asked, and that is worth knowing - record the section as `from: unanswered` - write, in the section itself, that it was asked and not answered **Never write `TBD`.** `TBD` means "someone will get to this", and `compass plan lint` scans for it as an unfinished placeholder. A section that was asked about and deliberately left open is *finished*. Two different states, and only one of them is true: > No non-goals were stated in the source, and none were supplied when asked. That sentence is a complete section. `TBD` is not. ## What you write | Where | What | |---|---| | `intent.md` | the reshaped document, following `templates/intent.md`'s sections | | `manifest.yml` → `intent_source.sections` | one entry per section: `name`, `from` (`source` \| `answer` \| `unanswered`), and `answer_id` where it cites one | | `manifest.yml` → `intent_source.interview` | every question asked, with its answer or `null` | `intent-source.md` is never edited. It is the record `intent.md` is checked against, and a reviewer who wants to know whether you invented something reads the two side by side. ## Hand-off > I have written `intent.md` from the source document at `<origin>`. > > N sections came from the source, M from questions I asked, K were asked > about and left open. The questions and answers are recorded in the manifest. > > Worth a read before the acceptance criteria are written. Specifically: > - **Anything I have put in your words that is not what you meant** - I > reorganized the source, and reorganizing can change emphasis. > - **The sections marked as left open** - they are deliberate, not missing, > and if that is wrong now is the cheap moment. > - **The first slice**, if one was agreed. Everything downstream follows it. Fill in the real origin and counts. A hand-off that still says `<origin>` has
From ayeo-io/compass, Apache-2.0.
More from ayeo-io/compass
- 0.584score
tdd-discipline
ayeo-io
The red-green-refactor cycle as Compass runs it, including what to do when a change has no natural failing test.
SkillAFreshness: A★ 1Claude Code +9 - 0.581score
adaptive-routing
ayeo-io
Reading the three scored dimensions for the CLI.
SkillAFreshness: A★ 1Claude Code +9 - 0.581score
bdd-specification
ayeo-io
Writing and checking Given/When/Then acceptance scenarios.
SkillAFreshness: A★ 1Claude Code +9 - 0.581score
behavior-mapping
ayeo-io
Capturing existing behavior as scenarios before changing it.
SkillAFreshness: A★ 1Claude Code +9 - 0.581score
compass
ayeo-io
Use Compass stages and roles to deliver a change or resume an issue, using the shared kit and instructions.
SkillAFreshness: A★ 1Claude Code +9 - 0.581score
flow-management
ayeo-io
Priorities across issues, blockers and the digest.
SkillAFreshness: A★ 1Claude Code +9 - 0.581score
governance-check
ayeo-io
Checking a design against the guardrails and strategies in force.
SkillAFreshness: A★ 1Claude Code +9 - 0.581score
receiving-code-review
ayeo-io
Answering review comments: check each against the code, push back with reasons, record the outcome.
SkillAFreshness: A★ 1Claude Code +9 - 0.581score
systematic-debugging
ayeo-io
Debugging from evidence: read the failure, form one hypothesis, test it.
SkillAFreshness: A★ 1Claude Code +9 - 0.581score
evidence-gates
ayeo-io
What counts as gate evidence, and how to clear a gate.
SkillAFreshness: A★ 1Claude Code +9 - 0.577score
compass-runtime
ayeo-io
Which command runs each stage, which agent owns it, where issue state lives.
SkillAFreshness: A★ 1Claude Code +9 - 0.577score
plan-authoring
ayeo-io
Writing a technical design and its self-review.
SkillAFreshness: A★ 1Claude Code +9
Signals
- Stars
- 1
- Forks
- 0
- Open issues
- 23
- Last commit (this folder)
- Oct 20, 2026 (9 days ago)
- Last release
- Oct 20, 2026 (9 days ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- Apache-2.0
- Repository
- github.com/ayeo-io/compass
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from ayeo-io
- 0.584score
tdd-discipline
ayeo-io
The red-green-refactor cycle as Compass runs it, including what to do when a change has no natural failing test.
SkillAFreshness: A★ 1Claude Code +9 - 0.581score
evidence-gates
ayeo-io
What counts as gate evidence, and how to clear a gate.
SkillAFreshness: A★ 1Claude Code +9 - 0.581score
behavior-mapping
ayeo-io
Capturing existing behavior as scenarios before changing it.
SkillAFreshness: A★ 1Claude Code +9 - 0.581score
adaptive-routing
ayeo-io
Reading the three scored dimensions for the CLI.
SkillAFreshness: A★ 1Claude Code +9 - 0.581score
bdd-specification
ayeo-io
Writing and checking Given/When/Then acceptance scenarios.
SkillAFreshness: A★ 1Claude Code +9 - 0.581score
compass
ayeo-io
Use Compass stages and roles to deliver a change or resume an issue, using the shared kit and instructions.
SkillAFreshness: A★ 1Claude Code +9
Cite this page
intent-interview skill. Agent Mart. https://agentmart.dev/skills/ayeo-io-intent-interview (last updated 2026-10-24)