Commands
Agile Work
Every official skill is small (under 60 lines), ends with an explicit "Done when" completion condition, stays provider-neutral, and ships with at least three eval scenarios, including prompts that must not trigger it.
Skills
| Skill | Pack | Activation | Requires | What it does |
|---|---|---|---|---|
diagnosing-bugs | core | auto | Reproduce, separate symptoms from the root cause, make the smallest fix that explains the evidence, verify the original failure is gone. | |
checking-blast-radius | core | auto | Find what a change could break elsewhere, and prove or disprove the one or two assumptions it depends on. Includes a checklist of surfaces where changes leak. | |
verifying-changes | core | auto | Produce the smallest meaningful proof that a changed behavior works, rather than running everything. | |
reviewing-changes | core | auto | git (network optional) | Review the actual diff for correctness, regressions, missing behavior, scope mistakes, and security where relevant, with a location and failure scenario for every finding. |
filing-pr | github | auto | git, gh, network | Check for an existing PR, follow title conventions, lead with the problem, include verification evidence, open the PR, return its URL. Includes the story ID when the work is tied to an Agile Work story. |
babysitting-pr | github | auto | git, gh, network | Monitor CI and review feedback, fix verified issues, keep scope, and stop at a defined done state (required checks green, feedback resolved, no merge blocker). |
resolving-conflicts | github | auto | git | Resolve merge, rebase, cherry-pick, or stash conflicts by understanding both sides' intent, then verify and finish the operation. |
interviewing-requirements | communication | manual | Ask a few high-value questions (after checking the repository first) until there is a shared, actionable plan. Uses structured questions when the provider offers them, plain text otherwise. | |
simplifying-explanations | communication | manual | Re-explain code, a concept, an error, or a previous answer at a simpler level. | |
creating-epics | agile | auto | Create or refine an outcome-level Agile Work epic: problem, outcome, scope, non-goals, success signals. No stories unless asked. | |
writing-stories | agile | auto | Break an epic or requirement into vertical-slice stories with observable acceptance criteria and only real dependencies. | |
working-story | agile | auto | Implement a story from its acceptance criteria through verification; check only verified criteria; end in the right status. | |
reviewing-story | agile | auto | Review an implementation against its story criterion by criterion, and refuse to close a story that is not delivered. |
Invoke manual skills by name, for example "use the interviewing-requirements skill and grill me about the onboarding redesign", /interviewing-requirements in Claude Code, or $interviewing-requirements in Codex.
Interaction contracts
Each skill declares capabilities.userInteraction, which decides whether the project's question preference applies to it:
userInteraction | Skills |
|---|---|
none (no overlay) | checking-blast-radius, verifying-changes, reviewing-changes, simplifying-explanations, reviewing-story |
optional | diagnosing-bugs, filing-pr, babysitting-pr, resolving-conflicts, creating-epics, writing-stories, working-story |
required | interviewing-requirements |
Packs
| Pack | Skills | Notes |
|---|---|---|
@agileflow/core | diagnosing-bugs, checking-blast-radius, verifying-changes, reviewing-changes | Everyday engineering workflows. Installed by agileflow init --yes. |
@agileflow/github | filing-pr, babysitting-pr, resolving-conflicts | Pull request workflows through gh or your provider's GitHub tooling. |
@agileflow/communication | interviewing-requirements, simplifying-explanations | Manually invoked interaction styles. |
@agileflow/agile | creating-epics, writing-stories, working-story, reviewing-story | Agile Work workflows. Offered by agileflow work init. |
Every pack entry uses a ^1 range.
agileflow add @agileflow/github