From taskforge-pro
Automatically and continuously executes all tasks in the current sprint. Use when the user says "/taskforge-execute-all", "run everything", "run the sprint", "keep going automatically", or similar. Supports wave-based parallel execution and runs through the sprint without stopping.
How this skill is triggered — by the user, by Claude, or both
Slash command
/taskforge-pro:taskforge-execute-allThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Continuously execute the remaining tasks in the current sprint without stopping.
Continuously execute the remaining tasks in the current sprint without stopping.
/taskforge-execute (project selection, approved plan, execution-state)_workspace/projects/{projectId}/Before starting execution, verify that the current sprint's tasks are real planned tasks:
project-plan.json and execution-state.jsonproject-plan.json with a valid plan fieldplan, skip it with a warning and continue with the remaining taskstelemetry.jsonl as a sprint_start eventThis prevents phantom tasks (tasks in execution-state that were never in the plan) from running silently.
If tasks have a wave field, execute in wave order:
Wave 1: [t1, t2, t3] — no dependencies, parallel execution
↓ (after all complete)
Wave 2: [t4, t5] — depends on Wave 1, parallel execution
↓ (after all complete)
Wave 3: [t6] — depends on Wave 2, single execution
wave field are executed sequentially in dependency orderFor each task, perform the same flow as /taskforge-execute:
Every completed task gets a handoff written to handoffs/{task-id}.json. This is no longer conditional.
Handoff content: what was built, key decisions made, anything the next task needs to know.
Track guardrail events across the sprint:
guardrailCount in memory during the sprint runsprint_complete telemetry eventguardrailCount >= 3, surface a warning in the sprint summary prompting the user to review /taskforge-statusAfter each task completes (same as /taskforge-execute):
TODO, FIXME, PLACEHOLDER, NOT IMPLEMENTED in new codenull, undefined, or falseconsole.error / throw paths that are unreachable (dead catch blocks)handoffs/{task-id}.json under silentErrorsguardrailCount by the number of issues foundAppend each event to telemetry.jsonl (newline-delimited JSON, one object per line).
{"t":"2026-04-19T10:23:00Z","event":"task_start","taskId":"M1-S1-T3","wave":1,"model":"sonnet","size":"feature","referencesInjected":["ui-pattern.html"],"cotUsed":true,"manifestFiles":5}
size: "tiny" | "normal" | "feature" | "milestone" (matches plan task size)referencesInjected: array of reference file paths injected into contextcotUsed: whether a CoT template from prompts/cot/ was injectedmanifestFiles: total number of files read from contextManifest{"t":"2026-04-19T10:41:00Z","event":"task_end","taskId":"M1-S1-T3","outcome":"pass","retryCount":0,"tokens":12450,"costUSD":0.42,"wallMin":18}
outcome: "pass" | "fail" | "skip"retryCount: number of retries before outcome (0 = first-try pass){"t":"2026-04-19T10:45:00Z","event":"sprint_complete","sprintId":"M1-S1","tasksCompleted":5,"tasksFailed":0,"totalCostUSD":1.2,"guardrailEvents":0}
Write to _workspace/projects/{projectId}/telemetry.jsonl. Create the file if it does not exist. Never overwrite — always append.
After all tasks in the sprint complete, display:
Template — fill from actual sprint state:
스프린트 {sprintId} "{sprintName}" 완료
Wave {n} (병렬): {taskName} ✅ {taskName} ✅
Wave {n} (병렬): {taskName} ✅ {taskName} ✅
Wave {n}: {taskName} ✅
소요: {duration} | 비용: ${cost} | 가드레일 발동: {n}회
다음 단계:
→ 스프린트 검증이 백그라운드에서 진행 중 (결과는 /taskforge-status에서 확인)
→ 다음 스프린트로 자동 진행 중...
If guardrailCount >= 3, append:
⚠️ 이번 스프린트에서 가드레일이 여러 번 발동됐습니다. /taskforge-status로 확인하세요.
At every sprint boundary:
/taskforge-validate sprint, async). It is advisory only and must NOT block the next sprint. Results land in validations/sprint-{id}.json with status: "advisory". Failures become warnings carried into the next milestone gate.This is the core change that makes wave parallelism work across sprint boundaries: the executor never stalls on sprint review.
After a sprint completes, automatically refresh the follow-up plan:
plan fields need updating based on what actually happened?Plan refresh happens automatically at sprint boundaries. No separate refresh command needed.
When the final sprint of a milestone completes, display:
마일스톤 M1 모든 스프린트 완료!
권장 순서:
1. /taskforge-validate milestone — 자동 품질 검증
2. /taskforge-playtest — 직접 플레이해서 확인
3. /taskforge-retro — 회고 + 다음 마일스톤 계획 업데이트
Then halt. Do not start the next milestone automatically. The user must run /taskforge-validate milestone first.
If the user says "stop", "halt", "pause", or similar mid-execution, stop after the current task completes.
Then use /taskforge-execute to continue one at a time, or /taskforge-execute-all to resume continuous execution.
/taskforge-validate milestone before starting the next milestone./taskforge-status and rolled into the milestone gate review.telemetry.jsonl is the source of truth for /taskforge-cost aggregation.Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub chodolmu/taskforge-pro --plugin taskforge-pro