scrum-task-update

Files (1)

• SKILL.md

SKILL.md (first lines)

Use the scrum-studio MCP. Reuse known project and task IDs. For a code use `task_get(project,code)`; for an ID use `task_get(project,id)` only if missing context is needed. If only a title is known, resolve the project with `projects` and use bounded task_search(project,query) to identify the task; clarify ambiguous matches before writing.

Task `code` (for example `WEB-1`) is the human reference, numbered from 1 per project. Resolve it to the returned internal numeric `id` within the intended project before updating; never assume its suffix is the numeric ID. Prefixes can be shared by projects.

- Call `task_update(project, id, ...changed fields)`; omit unchanged fields. Combine related edits into one call. Keep task text in English.
- `files` replaces the planned edit-path list; claim=true atomically reserves that plan. Check claimed=false before doing work. Respect project WIP; use [scrum-file-locks](../scrum-file-locks/SKILL.md) for discovered files/conflicts. A default-policy transition to Done needs fresh evidence and complete checklist; [scrum-complete](../scrum-complete/SKILL.md) covers that workflow. Leaving In Progress releases file reservations too.
- Category is editable text (max 80 characters, default General). Reuse known English category names. Optional `log` appends a concise authored work entry (max 2000 characters) in the same call. Use [scrum-activity](../scrum-activity/SKILL.md) for comments or history, not routine full-history reads.
- Starting In Progress requires `task_claim(project,id,actor)`, which assigns owner and returns a private token. Active locks require actor+token for every edit. Use `owner_id` (0 unassigns) to assign without locking, and release before reassignment. Read [scrum-users](../scrum-users/SKILL.md) for identity, renewal and recovery rules. Never bypass another worker's lock. Moving away from In Progress automatically releases it.
- To edit and start atomically, use `task_update(...,actor,claim=true,...fields)` and retain the returned token. Do not start execution until it succeeds. On Done the server resolves completed blockers and returns fully unblocked Blocked tasks to TODO with an authored comment; do not duplicate those actions. Include your actor ID when completing a task.
- Exact statuses: `TODO`, `In Progress`, `Blocked`, `In Review`, `Done`, `Cancelled`. Use In Progress when work starts, Blocked for an actual impediment, In Review for pending human review, Done for verified work, and Cancelled for authorized discarded work with cancel_reason. Cancellation releases locks, removes its dependent blocker links and returns fully unblocked Blocked tasks to TODO with authored comments; it does not count as successful delivery. See scrum-cancel only when needed. A request to move a record alone does not authorize doing the underlying work.
- Priority: 1 High, 2 Normal, 3 Low. Priority sorts tasks within the workflow's state order; it does not override In Progress before actionable Blocked before TODO.
- Blocked requires `blocked_reason` or dependencies. Prefer dependency links without duplicate prose. New prose alongside dependencies requires `blocked_reason_kind=dependency` (only describes linked prerequisites; clears when all resolve) or `external` (separate constraint; preserved). Mixed external/task reasons use external. Legacy prose defaults to external; inspect before reclassifying. Check evidence before clearing an external reason. Dependencies must all be Done before progressing; do not erase real dependencies to bypass this check.
- `checklist=[{text,done}]` and `dependencies=[id,...]` each replace their full list; retrieve the current task if it is not already known before making a partial list edit. Preserve unrelated entries. Use `[]` only when the user intends to clear the list. Check items only when completed; do not mark required unfinished work Done.
- Checklist completion does not automatically change task status. Moving status does not automatically complete its checklist. Keep these states accurate independently.
- The server records structured state intervals automatically; do not submit guessed durations or duplicate timer logs. Use task_timing only for a requested timing check; routine edits and renewals do not reset clocks.
- Report the tool's returned ID/status or error concisely. Do not repeatedly retry a validation error without correcting its cause. The MCP has no task delete operation.

For executing a sequence of tasks and repeatedly reassessing blockers, use [scrum-work](../scrum-work/SKILL.md).

Numeric IDs are local to a working database. After snapshot restore/import resolve code again and discard cached task IDs and tokens.

Reference files use task_attachment(project,id,action=list|add|get|remove). Task context and list/get default to metadata only; never fetch content automatically. add needs active actor, a plain allowlisted filename and explicit content_base64 (max5MiB decoded, ten per task); no local source path is accepted. Mutations respect active task actor/token and usage policy. get needs attachment_id; include_content=true explicitly loads base64 for a requested preview/download. remove needs actor/lease credentials and removes the reference plus its local blob. Keep references through normal task transitions and ownership changes. Supported image previews are PNG/JPEG/GIF/WebP; other accepted documents download only. Missing content and JSON export limitations are explicit: metadata is portable but .scrum/local/attachments must be transferred separately. Never treat a metadata reference as automatically loaded model context.

Execution progress, discoveries, current results and next steps belong in task_activity(kind=work|comment), or the log field on an already-required status update. Do not repeatedly rewrite description/acceptance_criteria with progress. Change those fields only for a genuine scope/acceptance change, recording the reason in an authored update.

Use task_get(version_seen) with the last returned version; changed=false means preserve cached task fields. For handoff use task_update(...,continuity="Result / files / verification / next step",actor,token), max2000. Authorship is logged. continuity_stale means task context changed after that summary; verify it before relying on it. scope_size=Unknown|Small|Large and scope_reason express estimated outcome complexity; Small/Large require a reason and preserve all safety gates.

Task model routing: fixed task model (human-only) overrides the human agent-profile model; otherwise automatic model_tier selects economy/standard/advanced. AI may set model_tier with concise model_reason during existing planning or task_update with normal ownership. Auto defaults to standard regardless of scope size. Size is not difficulty: choose economy only for mechanical low-risk work, standard for implementation/debugging/Unity integration, and advanced for demonstrated difficult reasoning. Set the tier and concise reason during existing planning or a necessary task update; never launch a separate selector model. No separate model call makes this decision. Manual model IDs are never replaced automatically. Human task_update(id,actor,model) alone can queue an override during an active claim, applying on the next launch without editing another worker’s code. Managed workers execute only their offered task, close, then exit; the supervisor re-evaluates the next task. For an automatic tier change during work, preserve task continuity and close safely; at most two routing relaunches occur. The desktop chat itself cannot change its own model through this field. No Claude account probe or paid test run is authorized by implementation.

Choose optional verification_level (auto/minimal/targeted/full) and verification_reason (max 500 characters; required for explicit levels) during ordinary task planning/create/update, without a separate model call. Auto adds no default compact AI context. Minimal uses justified inspection or manual checks for low-risk changes; targeted checks affected behavior and regressions; full runs broad regressions for shared contracts, storage or unresolved risk. These record intent, not successful evidence or permission to skip mandatory checks. If automated tests are skipped, record the reason and actual manual/inspection result. Acceptance criteria and required evidence still apply; scripts/verify.py keeps its conservative safe check selection and reuse rules.
Use optional `context_group` (single-line name, max80 characters; blank clears) in task_create/task_update or task_plan items for planned shared-context membership. Use the same name on related tasks; do not duplicate grouping instructions in descriptions. Task list > Group by > Context group displays this metadata. It grants no claims, reservations, dependency changes or execution authorization. Agent batch separately displays actual execution batches; compatibility, dependencies and fresh ownership still apply.

A human moving Blocked to TODO explicitly requests a retry: the server clears the previous blocker text, including unchanged text echoed by the web form, and records it in history. Dependency links, reservations, Stop, quota and native permission checks remain enforced. Agents must still resolve blockers explicitly; saving a different nonempty reason requires Blocked.

From martiin3d/scrumstudio, MIT.

More from martiin3d/scrumstudio