askill
he-workflow

he-workflowSafety 95Repository

Orchestrates a harness-engineered workflow across intake, spike, plan, implement, review, verify-release, learn, and doc-gardening using a slug-based docs/specs and docs/plans system. Use when running an initiative end-to-end with parallel subagents.

1 stars
1.2k downloads
Updated 2/16/2026

Package Files

Loading files...
SKILL.md

HE Workflow Orchestrator

Use this skill to run the full lifecycle with parallel execution and phase gates.

Inputs

  • Initiative request text, OR
  • Existing slug (YYYY-MM-DD-topic), OR
  • Existing active plan path (docs/plans/active/<slug>.md)

Source of Truth

  • Human intent: docs/specs/<slug>.md
  • Spike findings: docs/specs/<slug>-spike.md (if a spike was run)
  • Execution plan: docs/plans/active/<slug>.md (plan_mode: lightweight|execution)
  • Generated project context: docs/generated/ (for example docs/generated/db-schema.md)

Slug Rules

  • Format: YYYY-MM-DD-kebab-topic
  • Reuse one slug across spec and plan artifacts
  • Never create a second slug for the same initiative

Required Phase Order

  1. intake
  2. spike (optional — when feasibility or approach is unclear)
  3. plan
  4. implement
  5. review
  6. verify-release
  7. learn

doc-gardening is optional and periodic.

Re-entry Rules

When a gate fails or a phase reveals issues requiring earlier work:

  • NO-GO at verify-release: Return to he-implement for minor fixes or he-plan for design-level issues. Add a Decision Log entry explaining the loop-back.
  • Task failure during implement: Mark task as blocked in Task DAG, log the reason in Progress Log, continue with remaining batch tasks.
  • Fundamental design issue at review: Return to he-plan with a Decision Log entry. Re-validate the affected tasks before resuming implementation.
  • Every re-entry creates a Progress Log entry explaining the loop-back and the target phase.

Orchestration Steps

  1. Resolve slug.
  2. Validate workflow structure exists:
    • docs/specs/
    • docs/plans/active/
    • docs/plans/completed/
    • docs/generated/
    • docs/references/
  3. If any required directory is missing, run he-bootstrap before continuing.
  4. Ensure docs/specs/<slug>.md exists; if not, run he-intake.
  5. If spike_recommended: yes in the spec metadata, run he-spike before planning.
  6. Read plan_mode from docs/specs/<slug>.md (lightweight or execution).
  7. Ensure docs/plans/active/<slug>.md exists for the selected plan_mode; if missing, run he-plan.
  8. Validate every task/subtask has a Task Details section heading in format #### [ ] <task_seq> [Action-oriented title].
  9. For execution mode: validate plan tasks/subtasks include concrete files_to_change, tests_to_run, and verify_commands in Task Details.
  10. For lightweight mode: validate tasks include files, steps, verify, and done_when.
  11. Validate every task specifies test_type (unit or e2e) — no mocks.
  12. Validate implementation steps are markdown checkboxes.
  13. Validate Task DAG status reflects current completion state.
  14. Validate the plan has both Decision Log and Progress Log sections.
  15. Refresh generated context in docs/generated/ if stale.
  16. Run he-implement with DAG batches.
  17. Run he-review and enforce priority gate.
  18. Run he-verify-release.
  19. Run he-learn and archive plan.

Subagent Strategy

Use subagents liberally throughout the workflow to keep the orchestrator's context clean:

  • Intake: Subagents for codebase research and prior-work discovery.
  • Spike: One subagent per approach being compared.
  • Plan: Subagents to explore different codebase layers in parallel.
  • Implement: One subagent per task in each DAG batch — this is the primary parallelism lever.
  • Review: One subagent per review dimension (correctness, architecture, security, simplicity).
  • Doc-gardening: One subagent per scan target area.

The orchestrator delegates work to subagents and handles integration, plan updates, and gate enforcement in the main thread. Never do implementation or review work directly in the orchestrator context.

Parallel Execution Contract

  • Execute all dependency-free tasks in the same batch concurrently via subagents.
  • Use isolated worktrees per task whenever task blast radius is non-trivial.
  • Batch integration is required before advancing to the next dependency layer.
  • On conflict, split or sequence tasks and retry integration.

Branch and PR Convention

  • One branch per initiative slug.
  • Use worktree-per-task for parallel execution when blast radius is non-trivial.
  • PR is created at the implement-to-review boundary.
  • PR title follows: <type>(<scope>): <slug> <summary> (matching commit convention).
  • Commit message convention:
    • docs(intake): <slug> ...
    • docs(plan): <slug> ...
    • docs(review): <slug> ...
    • docs(verify-release): <slug> ...
    • docs(learn): <slug> ...

Gates

Doc Commit Gate (Hard)

Before phase transition:

git status --short docs
  • Required phase docs must be committed.
  • No uncommitted docs changes at the boundary.

Plan Specificity Gate

  • Each task/subtask in docs/plans/active/<slug>.md must include a heading in format #### [ ] <task_seq> [Action-oriented title].
  • For execution mode: each task must include concrete file paths, test commands, objective, implementation steps, risks, rollback impact, and evidence.
  • For lightweight mode: each task must include files, steps, verify, and done_when.
  • Every task must specify test_type (unit or e2e) — no mocks.
  • Step checkboxes (implementation_steps in execution mode, steps in lightweight mode) must be markdown checkboxes.
  • Task DAG status is the single source of truth for completion state.
  • Plans that use vague placeholders (for example ..., TBD) for files/tests cannot proceed to implement.
  • Plan must include explicit Decision Log and Progress Log.

Priority Gate

  • Unresolved critical or high priority findings block progression.
  • medium and low priorities may proceed only with explicit acceptance notes in the plan.

Generated Context Files

  • docs/generated/db-schema.md (if present)
  • docs/generated/api-schema.md (if present)
  • docs/generated/component-tree.md (if present)
  • docs/generated/dependency-graph.md (if present)
  • Other generated context files used as reference for planning/review/verification
  • Each generated file should include a last_updated timestamp

Completion

Workflow is complete when:

  • Plan is moved to docs/plans/completed/<slug>.md
  • Learn updates are written
  • Docs are committed

Transition Options

Present 2-3 explicit next-step options with a recommended default. Use request_user_input (Codex) or AskUserQuestion (Claude Code) in Plan mode; otherwise ask in chat. Wait for user selection before proceeding.

At least one option must name the next phase step (intake -> he-spike or he-plan, spike -> he-plan, plan -> he-implement, implement -> he-review, review -> he-verify-release, verify-release -> he-learn, learn -> he-doc-gardening or next-initiative he-intake).

Install

Download ZIP
Requires askill CLI v1.0+

AI Quality Score

92/100Analyzed 2/19/2026

Highly comprehensive workflow orchestrator skill with clear phase structure, detailed orchestration steps, subagent strategy, and gate enforcement. Well-organized with explicit inputs, source of truth, re-entry rules, and transition options. Tags provide good discoverability. Minor deduction for internal tooling references (he-bootstrap, he-intake) that limit portability, but the core methodology is reusable.

95
95
80
95
90

Metadata

Licenseunknown
Version-
Updated2/16/2026
Publishermattjefferson

Tags

apigithub-actionsllmsecuritytesting