بنقرة واحدة
workflow-execute
Session-based work execution with progress tracking and knowledge capture
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
القائمة
Session-based work execution with progress tracking and knowledge capture
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Backlog-scale orchestration on top of /workflow — drives PM items through refine → plan → implement → review → local-merge via role-specialized sub-agents dispatched in parallel waves, each item isolated in its own git worktree. Bare /swarm summarizes state; /swarm <goal> runs the orchestrator; /swarm:setup sets up the charter and .agent-tools/ umbrella; /swarm:continue resumes a paused run.
Capture durable knowledge from any engineering work — debugging solutions, refactors, features, design decisions, reusable patterns — and maintain memory quality so each unit of work compounds the next
Principal workflow entry — resolve portfolio mode (swarm resume/handoff or unit state machine), drive work without inventing path; hard-stop when path is not established
Initialize and maintain planning root (.agent-tools/planning preferred), conventions (tracks, gates, integration), runs ledger scaffold, and shared memory under .agent-tools/memory/ (AGENTS.md memory-link) for the workflow family.
Parent skill for the workflow family — principal session entry is /workflow:continue (also bare /workflow). Horizon mapping, brainstorm, refine, plan, execute, review, audit, compound. Vertical-slice and deliverable-partition modes.
Detect gaps in the canonical skill corpus, propose targeted improvements, validate changes, and present for review. Inventories src/** skill trees; optional run-ledger seeds for process IP gaps.
| name | workflow:execute |
| description | Session-based work execution with progress tracking and knowledge capture |
| argument-hint | [--worktree] [planning directory, plan file, 'continue', or session summary text] |
| user-invocable | true |
Execute work plans while maintaining session continuity and capturing knowledge.
$ARGUMENTS
Before interpreting input, extract the --worktree flag if present:
--worktree in $ARGUMENTSWORKTREE_MODE=true and strip --worktree from $ARGUMENTS before Input InterpretationWORKTREE_MODE=falseValidation (stop with error if any fail):
--worktree continue is invalid: Creating a new worktree for an existing session is nonsensical. Error:
ERROR: --worktree cannot be used with 'continue'.
To resume work in an existing worktree, cd to the worktree directory and run:
/workflow:execute continue
git rev-parse --show-toplevel differs from the main repo root (CWD is already a worktree), error:
ERROR: Already inside a git worktree. Cannot nest worktrees.
Run /workflow:execute without --worktree from this worktree.
--worktree is passed AND session-state.md already has a worktree: path, error:
ERROR: A worktree is already recorded in session-state.md: <path>
Run /workflow:execute without --worktree — it will detect and enter the existing worktree automatically.
| Input Pattern | Interpretation |
|---|---|
./planning/<project>/ | Start/continue project from planning subdirectory |
./planning/ or empty | Auto-detect from ./planning/*.md or ./planning/*/ |
*.md file path | Execute specific plan file |
continue | Resume last active session |
LIN-[0-9]+ or [A-Z]+-[0-9]+ | Fetch via Issue Retrieval Strategy, then direct execution |
| Text summary | Steering input for next session |
For simple tasks that don't need full planning docs (clear bugs, small enhancements, issue-as-plan).
Also the process used when /workflow:continue classifies track micro.
references/direct-issue-execution.md (fetch issue, confirm, execute, quick review, integrate, compound disposition)/workflow:plan or refine; resume via continuePM retrieval: @workflow (planning/pm-integration.md). Tracks: @workflow references/tracks.md.
Precedence:
ls ./planning/*/session-state.md 2>/dev/null
ls ./planning/*.md 2>/dev/null
Support project subdirs (planning/<project>/) and flat planning/ files. Both modes use
implementation-plan.md + session-state.md; requirements.md is file mode only.
Optional visual_plan: in session-state is approval metadata only — execute always follows
the markdown implementation-plan.md, never a visual plan URL or MDX folder.
If session-state.md has worktree: or WORKTREE_MODE=true, load and follow
references/worktree-enter.md (detect existing, create via @git worktree-create, rename branch,
dependency restore via dependency-establishment.md, validate planning docs committed, record path).
Surface the ERROR strings defined there (including planning-docs-missing-in-worktree).
session_count: 1requirements.md when work_item / PM source is setbranch: in session-state; never start implementation on main/master<type>/<ISSUE-KEY> or short description)IMPORTANT: Never begin implementation work on main/master. Always create or switch to a working branch first.
Show project, session #, progress, current focus, last session summary.
Next incomplete plan task; honor steering; parallelize with sub-agents when safe; confirm if unclear.
Quality detail: load quality-checkpoints.md. Layer order: @workflow
(references/decomposition-modes.md). Testing strategy: @test-strategy. Logging: logging.md.
while (tasks remain):
1. Mark task in_progress in TodoWrite
2. Read relevant files from plan
3. Look for existing patterns in codebase
4. Write tests for next behavior — see @test-strategy
5. Implement minimal code to pass (Green)
6. Refactor if needed, run full tests
7. Mark task completed in TodoWrite
8. Update plan file ([ ] -> [x])
9. Check for story/slice completion
10. Check for session boundary
CRITICAL: Commit each decomposition unit independently (story/slice or sub-issue/deliverable).
For each unit:
feat(scope): description (ISSUE-ID)Anti-patterns: one big commit at the end; closing a sub-issue with a paraphrased AC instead of the verbatim parent AC.
Before marking a task complete:
Run project-appropriate checks (pytest/mypy/ruff, npm test/tsc/eslint, etc.). Full checklists:
quality-checkpoints.md.
Update plan checkboxes and session-state.md after significant milestones.
Before stopping or generating a handoff:
Finish remaining work if possible; else handoff; if blocked, document in session state then handoff.
Slice complete; milestone; user ending; context stale.
./planning/<project>/session-state.md using
templates/session-state.md (load and fill)templates/session-learnings.md; on Yes run /workflow:compoundtemplates/session-complete.md (includes worktree status/cleanup rules)Do not remove worktrees during handoff — user-initiated after all parallel sessions finish
(see template + references/worktree-enter.md + @git worktree-delete).
Stop; document; re-plan with /workflow:plan; resume from revised plan.
Document; create resolution task; parallelize if possible; escalate if critical path.
Read session-state, git log, implementation-plan; ask user if needed.
quality-checkpoints.md › Decision-Reconciliation at Close)quality-checkpoints.md, dependency-establishment.md, logging.md