Zum Inhalt

sprint-plan

Erstellt eine neue Sprint-Datei unter project/sprints/ mit Value-Statement, Features und value-verifizierendem Akzeptanzkriterium.

Create a new sprint file under project/sprints/ per the project sprint spec. Invoke when the user asks to plan a sprint, open a sprint, draft the next sprint, schedule a sprint, or pull roadmap items into a sprint. Also handles equivalent German-language requests. Resolves the next monotonic sprint number, walks the value-delivery contract (rejects operator-internal verbs in value_statement), pulls in roadmap items whose target_sprint matches, populates the features list (delegating to feature-decompose when decompositions are missing), names the value-verifying feature, and writes the file with status: planned. Supports resume on re-invocation per spec/claude/resumable-work/.

Anwenden wenn

  • you want to plan or open a new sprint
  • you want to draft the next sprint with roadmap items pulled in
  • you want to schedule a sprint and populate the features list

Nicht anwenden wenn

Siehe auch

Referenziert von

Sprint Plan

Creates project/sprints/<NNNN>-<slug>.md per spec/project/sprint/<canonical_language>.md. This skill owns the planned state of the sprint lifecycle: it resolves the sprint number, captures the value_statement from the operator, pins the value-verifying feature, and writes the planned-state markdown. It deliberately doesn't activate the sprint or move features—sprint-execute owns those transitions.

Why this is a skill, not an agent

  • Mid-flow user dialogue is core — the value statement is rejected if it begins with operator-internal verbs (refactor, migrate, …); resolving a rejection means a back-and-forth with the user, which an agent's structured-report shape can't carry.
  • Output flows back into the main conversation — the drafted sprint markdown, the resolved roadmap-item list, and the named value-verifying feature all live in the user's working context; isolating them behind an agent boundary would obscure the iterative review.
  • Orchestrator that may dispatch other skills — when a roadmap item lacks a feature decomposition, this skill reports the gap (and may chain feature-decompose per the skill-vs-agent hybrid pattern); orchestrators default to skill form.
  • Counter-dimension considered: a tool-restricted agent could perform the next-number resolution and the YAML write cleanly, but the load-bearing dimension is the user-facing rejection / iteration loop on value_statement — skill wins.

User-language policy

Detect the user's language and respond in it. The sprint markdown itself is written in the project's primary language (typically EN); the value_statement, ## Goal, and other body content stay in the project's primary language so consuming skills (sprint-execute, sprint-review) read a single deterministic surface. Frontmatter keys, lifecycle tokens, and feature / roadmap IDs are always English.

Preconditions

Before drafting any file, confirm:

  • Current working directory is inside a git repository whose project/ planning suite is adopted (at minimum project/roadmap.md and project/goals.md exist per spec/project/roadmap/). If project/ is missing entirely, stop and report—planning hasn't been bootstrapped.
  • A project/sprints/ directory exists or can be created. List the existing sprint files (ls project/sprints/*.md or equivalent) to read every present number from frontmatter.
  • The user has supplied (or will supply on prompt) a one-sentence value_statement from the end-user perspective. If they haven't, ask before proceeding; never invent it.
  • At least one roadmap item is proposed or active and either has target_sprint matching the next sprint number or the user has named which roadmap items the new sprint should pull in. An empty roadmap is a hard stop—report and hand back.

Operations

1. Resolve the next sprint number

Read every project/sprints/*.md file's frontmatter number field. The next sprint's number is max(existing numbers) + 1, or 1 when no sprint files exist. Sprint numbers are strictly monotonic across the project's lifetime per spec/project/sprint/ §Directory layout and file shape; never reuse a number from a cancelled or closed sprint.

Format the number as four-digit zero-padded (0001, 0007, 0042).

2. Capture and validate the value_statement

Ask the user for a one-sentence value statement phrased from the end-user perspective. Reject the statement (don't write the file; surface the violation with the offending verb quoted) when it begins with or is built around an operator-internal verb in either the project's primary language or German:

  • English: refactor, restructure, set up, configure, clean up, migrate, bump, update dependency
  • German: refaktorieren, umbauen, einrichten, konfigurieren, aufräumen, migrieren, aktualisieren, Abhängigkeit erneuern

The list MAY be widened per project via a sprint_rejection_verbs: key in the shared .github/release-skill-layer.yml override surface (the same file the sibling release-artifact and release-skill-layer specs use); a project MUST NOT introduce a separate .github/sprint-rejection-rules.yml for this purpose. The check is heuristic: when the user genuinely delivers an end-user-facing change whose phrasing happens to start with one of these verbs, they MAY override with a one-line rationale that is then recorded verbatim in the sprint's ## Goal section.

3. Derive the slug

Slugify the value_statement (or a shortened form the user confirms) into ASCII kebab-case: lowercase, hyphens between words, no punctuation, target ≤6 words. The slug SHOULD stay aligned with the value statement so directory listings remain self-describing.

Filename: project/sprints/<NNNN>-<slug>.md.

4. Resolve roadmap items

Read project/roadmap.md and collect every roadmap item whose target_sprint equals the new sprint's number. Present the list to the user and confirm. The user MAY add or remove items at this step; record the final list as roadmap_items in the sprint frontmatter.

An empty roadmap_items list is permitted only for hardening sprints whose features still link to a roadmap item per the feature spec. When the list is empty, the ## Goal section MUST state the hardening intent explicitly and cite the roadmap item(s) the features link to (per spec/project/sprint/ §Roadmap and feature linkage).

5. Resolve features

For each roadmap item collected in step 4, walk project/features/ and find features whose frontmatter roadmap_item matches and whose sprint is null or already equals the new sprint's number. Two outcomes per roadmap item:

  • Features exist — collect their id values into the features list candidate.
  • No features exist — surface the gap to the user and offer to dispatch feature-decompose (the skill that decomposes a roadmap item into one or more features per spec/project/feature/). Don't fabricate features inline; if the user declines decomposition, drop the roadmap item back out of the sprint and move on.

After confirming the final feature list with the user, set each chosen feature's sprint frontmatter field to the new sprint number in the same write operation that adds the feature's id to the sprint's features list. The bidirectional invariant (feature side ↔ sprint side) is the canonical check per spec/project/feature/ §Frontmatter schema; partial updates are forbidden.

Flip draft → ready under the gate. Setting feature.sprint = N is the moment a draft feature becomes schedulable, so this skill is the canonical write authority for the draft → ready transition per spec/project/feature/ §Lifecycle and gates. For every feature pulled in whose status is draft, before flipping it to ready verify the full gate: a non-empty ## Description, at least one **acceptance-<n>** criterion bullet, a populated consistency_check frontmatter object whose findings array is non-empty, and a populated ## Consistency notes section. If any gate condition fails, don't flip the status and don't pull the feature in — surface the missing precondition and route the operator back to feature-decompose to complete the decomposition (the consistency check in particular is feature-decompose's territory, never something to fabricate here). Flip status: draft → ready only on features that clear the gate, in the same write that assigns sprint. A feature already in ready (or any later state) keeps its status; only draft features are flipped here.

6. Name the value-verifying feature

Exactly one feature in the sprint MUST carry a non-null verifies_sprint_value: acceptance-<n> frontmatter field per spec/project/sprint/ §Value-delivery contract. Walk the candidate features:

  • If exactly one feature already declares verifies_sprint_value and its named acceptance criterion plausibly verifies the sprint's value_statement, confirm with the user and keep it.
  • If zero features declare it, ask the user to pick the verifying feature plus the specific acceptance-<n> identifier on that feature, and write the field. The feature-decompose skill is the canonical write authority at decomposition time; sprint-plan is the canonical write authority during planning per spec/project/feature/ §Frontmatter schema.
  • If more than one feature declares it, the constraint is violated; stop and ask the user to resolve down to exactly one before continuing.

7. Render the sprint file

Write the file with this exact shape (frontmatter keys in the declared order from spec/project/sprint/ §Frontmatter schema):

```text

number: status: planned started: null ended: null value_statement: artifact_ref: null last_commit: null roadmap_items: [, ] features: [, ]

Goal

Features

  • — status: ready
  • — status: ready

Out of scope

  • None>

Review notes

Populated by sprint-review at closure. ```

The ## Features body bullet list MUST mirror the features frontmatter list exactly; sprint-execute enforces the invariant on every later mutation, but the initial write must already be consistent.

8. Present and confirm

Diff the planned file content back to the user before writing. Confirm:

  • The resolved sprint number and filename.
  • The accepted value_statement (and any rejection-override rationale).
  • The roadmap_items and features lists, with the named verifying feature highlighted.

Only write the file once the user approves. Report back: the path written, the sprint number, the named verifying feature's id plus acceptance-<n>.

Gotchas

  • value_statement must avoid operator-internal verbs. "Refactor the release pipeline" reads as internal work; "Operators run a single command to ship a release" is the user-visible value the spec demands. The skill's heuristic flags imperative-form verbs that don't name a user benefit and asks the operator to rephrase before write.
  • The ## Features body list and the features frontmatter array MUST stay in lockstep. sprint-execute enforces the bidirectional invariant later, but this skill creates the initial pair; an asymmetric write (only frontmatter, only body) leaves the sprint malformed and blocks subsequent sprint-execute runs.
  • verifies_sprint_value lives on a feature, not on the sprint. The sprint frontmatter names which feature carries the verifier (verifies_sprint_value: F-<n>:acceptance-<m>); the feature carries the actual verifies_sprint_value: acceptance-<m> field per the feature spec. Authoring the verifier in the sprint frontmatter without setting it on the feature side leaves the cross-reference half-broken.
  • Sprint numbers are monotonic, never reused. The skill resolves the next sprint number by reading the highest existing <NNNN> under project/sprints/ plus the highest deleted number from git log -- project/sprints/. A retired sprint number isn't available for reuse, even after a cancelled sprint.
  • At most one sprint is active at a time. Creating a planned sprint while another is active is fine (planned sprints are queue items, not active commitments). The at-most-one invariant only applies to status: active; this skill writes status: planned and lets sprint-execute perform the planned → active promotion when the operator starts the first feature.

Examples

  • Read examples/01-create-sprint-from-roadmap.md when creating the first sprint by selecting features from the roadmap queue.
  • Read examples/02-reject-operator-internal-value.md when the user proposes a value_statement that starts with an operator-internal verb and the skill must refuse.
  • Read examples/03-dispatch-feature-decompose.md when a roadmap item lacks a feature file and the skill dispatches feature-decompose before continuing.

Resumability

Per spec/claude/resumable-work/, this skill is resumable: true. State is persisted to .resume/sprint-plan/<run-id>.yml after every successful user-approval gate and after each named phase boundary. On re-invocation, scan that directory for files with status: in_progress whose inputs: snapshot matches the current invocation; if one matches, prompt the operator with Resume run <run_id> from phase <phase> (last checkpoint <last_checkpoint_at>)? [resume / start-new / discard]. The state-file envelope (schema_version, run_id, inputs, phase, decisions[], status, ...) and the fail-closed semantics on schema or YAML errors are load-bearing in the spec; don't duplicate those rules here.

Hard rules

  • Never reuse a sprint number from a cancelled or closed sprint. Numbers are strictly monotonic per spec/project/sprint/ §Directory layout and file shape.
  • Never write a sprint with status other than planned. Activation is sprint-execute's authority.
  • Never flip a feature draft → ready without verifying the full §Lifecycle gate (non-empty ## Description, ≥ 1 acceptance-<n> bullet, populated consistency_check with a non-empty findings array, populated ## Consistency notes). A draft feature that fails the gate is reported as a gap and routed to feature-decompose; it is never pulled into the sprint nor silently flipped.
  • Never invent roadmap items, features, or audiences inline. Missing decompositions are reported as gaps; missing audiences are a roadmap-side concern, not a sprint-side fix.
  • Never persist a value_statement that begins with an operator-internal verb without an explicit user-supplied override rationale recorded in ## Goal.
  • Never write a features frontmatter list that diverges from the ## Features body bullet list at initial creation.
  • Never set verifies_sprint_value on more than one feature in the sprint, or leave it unset on every feature when the sprint is intended to close eventually.
  • Never edit any sprint other than the one being planned in this run; lifecycle transitions on existing sprints are sprint-execute and sprint-review territory.
  • When spec/project/sprint/ disagrees with this skill, the spec wins. Propose updating this skill rather than silently diverging.