flow-execution
Skill19 starsChanged 5 days agoApache-2.0View on GitHub
Use when executing a single Flow chapter's task blueprints in isolated subagent turns with pre-review Markdown sync (/flow:implement) or reporting chapter and task status (/flow:status).
In this repository10 other files it gives its agents.
- AGENTS.mdAGENTS.md
- debloatSkill
- flow-completionSkill
- flow-docsSkill
- flow-learnSkill
- flow-planningSkill
- flow-setupSkill
- flowSkill
- multi-harness-pluginsSkill
- okfSkill
The file
We split this skill into the rules it gives an agent
7 rules. About 0 are common, 0 are distinctive, and 7 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
7
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 13
Use for implement (../flow/references/implement.md) and status (../flow/references/status.md).
- Very specific: line 34
Read root .agents/bundles/specs/<flow_id>/spec.md, chapters/index.md (when has_chapters: true), and only the active current_chapter's …
- Very specific: line 35
Verify depends_on_chapters are completed with non-null last_reviewed_commit and planning_status == "complete".
- Very specific: line 36
Run the task preflight (dependencies_closed, worksheet_complete, verification_strategy_declared, plan_identity_matches, …
- Very specific: line 41
Chapter Recovery Checkpoint: Run the chapter's aggregate lint and test suite (exit 0).
- Very specific: line 49
Always run present-tense Markdown/OKF sync before Zero-Context Chapter Review so reviewers verify both code and living docs, and stamp …
- Very specific: line 50
Never mark a task or chapter completed without fresh exit 0 verification, and never create/mutate Git tags or push without explicit user …
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Flow Execution (Single-Chapter Subagent Pipeline, Pre-Review Sync, & Public-Seam TDD)
<!-- lifecycle-ownership: owner=flow-execution; operations=implement,status -->Trigger
Use for implement (../flow/references/implement.md) and status (../flow/references/status.md).
<!-- flow-execution-policy: start -->contract: worksheet-execution-v1
invariants:
- one-chapter-per-implement-run
- worksheet-first-bounded-read-set
- execution-preflight
- public-seam-tdd-no-mocked-sut
- fail-closed-no-production-mutation
- atomic-task-commit
- pre-review-present-tense-markdown-sync
- chapter-code-and-doc-review-before-stop
authority: skills/flow/references/implement.md<!-- flow-execution-policy: end -->Workflow
- Active Chapter Resolution & Execution Preflight (
/flow:statusor/flow:implement):- Read root
.agents/bundles/specs/<flow_id>/spec.md,chapters/index.md(whenhas_chapters: true), and only the activecurrent_chapter'sspec.md+tasks/*.md(../flow/references/state.md). - Verify
depends_on_chaptersarecompletedwith non-nulllast_reviewed_commitandplanning_status == "complete"; ifplanning_status == "empirical_dependent", stop cleanly with zero production edits and route to/flow:revise <flow_id> <chapter_id>. - Run the task preflight (
dependencies_closed,worksheet_complete,verification_strategy_declared,plan_identity_matches,target_symbols_exist). On mismatch, mark the taskstate: blocked([!]) with zero production edits and route to/flow:revise.
- Read root
- Isolated Subagent Task Pipeline (
1–4Tasks in Active Chapter Only):- Execute strictly one chapter (
1–4tasks) per/flow:implementrun, dispatching one fresh isolated@executorsubagent per task worksheet (model: inherit;run_workflowpipeline()withtype_name="self"andagents/executor.mdon Google Antigravity). - Each
@executorstays inside its worksheet'sBounded Read Manifest(read_set <= 6files), collects initial evidence (behavior_tdd/regression_tddRed first), implements through the real System Under Test (applying the Hypothesis-Driven Bug Diagnosis Loop on non-obvious failures), verifies Green + fast lint (exit 0), and atomically commits its owned files plus synced taskstate/commitand[x] [<sha>]checklist marker.
- Execute strictly one chapter (
- Chapter Closeout Sequence (Checkpoint → Present-Tense Sync → Zero-Context Review → Hard Stop):
- Chapter Recovery Checkpoint: Run the chapter's aggregate lint and test suite (
exit 0). - Present-Tense Markdown, OKF, & Full-Plan Sync: Dispatch an isolated sync subagent (
type_name="self"on Antigravity) to update affected.agents/bundles/knowledge/**and userdocs/**in present tense, rewritechapters/<ch_id>/spec.md,chapters/index.md, and rootspecs/<flow_id>/spec.md([k/N files documented]and## 6. Current State & Remaining Plan Snapshot) in place with zero chronological history (logging only in.agents/bundles/log.md), and commit the synced docs. - Zero-Context Chapter Code & Markdown-Sync Review: Dispatch isolated
@code-reviewer+@quality-reviewer(type_name="self"on Antigravity) acrosschapter_base_commit..HEADcovering both code/tests and synced Markdown/OKF files; remediate anyADOPTfindings and re-verify. - Review Stamp, Pointer Advance, & Hard Stop: Stamp
last_reviewed_commit: <REVIEWED_HEAD_SHA>andstate: completedinchapters/<ch_id>/spec.md, mark- [x] [<REVIEWED_HEAD_SHA>] <ch_id>and advancecurrent_chapter+last_reviewed_chapter+last_reviewed_commitin rootspecs/<flow_id>/spec.md, commit closeout metadata, and stop cleanly at the chapter boundary (recommending/compactor a fresh session before the next chapter).
- Chapter Recovery Checkpoint: Run the chapter's aggregate lint and test suite (
Guardrails
- Execute one task worksheet per subagent turn and at most one chapter (
1–4tasks) per/flow:implementinvocation; never chain multiple chapters in one session turn. - Always run present-tense Markdown/OKF sync before Zero-Context Chapter Review so reviewers verify both code and living docs, and stamp
last_reviewed_commitbefore stopping so cross-harnessStophooks pass cleanly. - Never mark a task or chapter
completedwithout freshexit 0verification, and never create/mutate Git tags or push without explicit user instruction.
Output
Return the completed chapter/task summary, verification exit codes, atomic task and sync commit SHAs, Chapter Review Adjudication summary, last_reviewed_commit, remaining chapters from the full plan, and /compact + next /flow:implement <flow_id> (or /flow:finish <flow_id>) handoff.
Validation
Confirm all chapter verification commands exited 0, every tasks/<id>-<slug>.md state/commit matches its spec.md checklist, root spec.md + chapters/index.md + knowledge/** + docs/** are synced in present tense without historical prose, and last_reviewed_commit covers every completed task commit in the chapter.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

