بنقرة واحدة
graphflow
Create, edit, and run persistent Codex workflow graphs with optional GitHub delivery.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
القائمة
Create, edit, and run persistent Codex workflow graphs with optional GitHub delivery.
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
Bootstrap software repositories from live evidence. Use for initial setup, inherited-codebase rehabilitation, repository mapping, conventions, documentation architecture, safe restructuring, or agent-ready tooling.
Solve ambiguous, high-stakes, or multi-variable questions with a labelled problem tree. Use for open-ended life, business, product, strategy, and root-cause reasoning; route concrete feature briefs to $grill-me and specific failures to $debugging.
Diagnose software failures through evidence-based root-cause analysis. Use for bugs, failing tests, runtime errors, data anomalies, stuck jobs, environment failures, and intermittent or expensive-to-reproduce incidents.
Turn a rough software feature request into one codebase-grounded implementation brief. Use for feature pitches, vague product requirements, solution exploration, or explicit “grill me” requests before coding; use $brainstorming for non-implementation questions.
Implement an agreed story, plan, or failing-test target and prove the resulting behavior. Use when an accepted contract authorizes production changes; route diagnosis-only work to $debugging and test design to $tdd.
Land finished work in a Git repository through Gate, Record, Commit, and Publish. Use when the user asks to close completed work, commit and push it, or open a pull request; use repository review workflows for reviewing an existing PR.
| name | graphflow |
| description | Create, edit, and run persistent Codex workflow graphs with optional GitHub delivery. |
| disable-model-invocation | true |
Its bundled runtime stores state under .codex/ and dispatches agent nodes through Codex. The runner
owns scheduling, recovery, node execution, and
durable state; the coordinator owns graph design, authority, integration, acceptance, and delivery.
Read automation-lifecycle.md. Create an unused stable
workflow_id, or require the exact ID to activate or resume an existing workflow:
python3 <skill-dir>/scripts/workflow_state.py resolve .codex/workflows --workflow-id <id>
Effective authority is the intersection of the user request, executor policy, and repository policy. Record commit, push, pull request, merge, deploy, destructive, network, and credential capabilities separately. Complete this phase when identity, objective, authority, and lifecycle state are unambiguous.
If delivery is required, read delivery-contract.md now. Prove the
configured adapter, hosting repository, remote, hostname, and available authentication route before
any executor dispatch; an unsupported provider or unavailable authorized preflight is a
waiting_external terminal, not a late Publish discovery.
For an existing noncanonical flow, read flow-editing.md. Produce a digest-bound reframe proposal and wait for approval before conversion. A canonical flow proceeds through the remaining gates.
Record observable objective, atomic requirements and acceptance, constraints, non-goals, authority,
and pivotal unknowns. Route a genuinely fuzzy root through the installed brainstorming skill; route
a concrete feature brief through grill-me. Missing required skills are design blockers under
skill-routing.md.
Read question-triage.md. Resolve pivotal contract unknowns before
ready, use evidence-backed reversible defaults for local uncertainty, and isolate branch-scoped
questions. Read preflight-costing.md only for a hard budget,
requested options, or material execution risk.
Complete this phase when question_gate is locked, clear, and has no unresolved pivotal question.
Read prototype-gate.md. Use the cheapest credible artifact: wireframe for subjective intent, isolated prototype for integration, characterization test for behavior, or dry-run for deterministic automation. Bind approval to the artifact digest; treat static or mocked evidence as intent proof only.
Complete this phase when the approved digest and limitations are recorded, or a deterministic exemption has a concrete machine-checkable reason.
Read graph-contract.md,
executor-runtime.md,
workspace-trust.md,
checkout-guard.md, and
method-routing.md. Copy
workflow-template to .codex/workflows/<id>/, then adapt every normative
artifact.
python3 <skill-dir>/scripts/validate_graph.py <graph.json> --phase draft
python3 <skill-dir>/scripts/question_gate.py lock <workflow-dir>
python3 <skill-dir>/scripts/workspace_manager.py init <workflow-dir>
python3 <skill-dir>/scripts/checkout_guard.py --repo-root <repo> init <workflow-dir>
Read integrity-contract.md, adapt requirement checks and independent verifier roles, then lock the oracle:
python3 <skill-dir>/scripts/evidence_runner.py lock <workflow-dir> --repo-root <repo>
python3 <skill-dir>/scripts/validate_graph.py <graph.json> --phase executable --ready
python3 <skill-dir>/scripts/evidence_runner.py validate <workflow-dir> --phase active --repo-root <repo>
Read shared-memory.md and initialize bounded coordinator-owned memory. Complete compilation only when executable validation passes and all executor, question, intent, workspace, memory, and oracle locks match the same semantic graph.
Read skill-routing.md and subagent-routing.md. Derive node skills from operations, pass only the node contract and named artifacts, and use the cheapest sufficient executor. Read dashboard.md only when a local read-only projection is useful.
python3 <skill-dir>/scripts/run_workflow.py <workflow-dir> --repo-root <repo>
Let the runner acquire the lease, reconcile durable state, check the primary checkout, execute the ready frontier, validate result envelopes, run acceptance checks, and continue to a terminal or waiting state. Read runtime-decomposition.md when an agent returns a structural decomposition proposal. Preserve failed coverage and retry once only after a classified contract repair.
Execution is complete when the runner reaches an honest terminal or digest-bound waiting state with no unaccounted live executor.
Give a fresh verifier the requirements, approved baseline, locked checks, integrated artifacts, and
known limits. Read evidence-calibration.md and record each claim
as verified, observed, inferred, or unverified.
python3 <skill-dir>/scripts/evidence_runner.py validate <workflow-dir> --phase complete --repo-root <repo>
python3 <skill-dir>/scripts/memory_state.py validate <workflow-dir> --phase complete --check-artifacts
Every required claim needs a current critical attestation; aggregate score cannot compensate for a failed requirement.
When delivery is authorized, read delivery-contract.md and the
installed ship skill. Preserve Gate → Record → Commit → Publish, discover the remote and base
branch from repository evidence, and bind one approval to the exact verified tree. Merge and deploy
remain separate authority.
Finish with outcome or PR, requirement-level evidence, changed scope, calibrated limitations, cost variance, and deferred optional work. Clear owned leases and processes while retaining evidence needed for recovery or review.