| name | execute-plan |
| description | HADES subagent-driven plan execution: stage reality-check, per-task
dispatch loop with review gates, no-tech-debt doctrine, hard gates per commit.
Use when operator invokes /hades:execute-plan or when implementing a stage plan.
|
| license | Proprietary |
| agentskills_version | 1 |
| keywords | ["execute-plan","subagent-dispatch","review-gates","stage","no-tech-debt","hades"] |
HADES — execute-plan skill (subagent-driven plan execution)
This skill provides HADES's plan execution methodology. Triggered by the
/hades:execute-plan slash command which loads superpowers:executing-plans +
superpowers:subagent-driven-development.
When to use
- Operator invokes
/hades:execute-plan <plan-path>
- When implementing a stage plan via subagent dispatch
Workflow
1. Load execution skills explicitly
skill_load("superpowers:executing-plans")
skill_load("superpowers:subagent-driven-development")
2. stage reality-check (BEFORE first dispatch)
Per feedback_plan_template_drift.md:
- Extract package.Symbol references from plan-file
- Verify each against current codebase
- Decision tree: 0-2 missing → note in prompt; 3-5 → fix plan first; ≥6 → halt
3. Per-task execution loop
For each task:
- Mark in_progress (TaskUpdate)
- Dispatch implementer subagent (foreground, not background)
- Wait for result
- Dispatch spec compliance reviewer
- Dispatch code quality reviewer (
superpowers:code-reviewer)
- Fix reviewer findings (no tech debt, no defer)
- Mark complete (TaskUpdate)
4. Doctrine — no tech debt, no defer
Reviewer "Minor" tag is severity, NOT permission to skip. Every Minor surfacing
real coverage gap MUST be fixed before next task.
5. Hard gates per task commit
make build && make lint && make test && make verify-invariants
go test -race ./... -count=2
GOOS=linux go build ./...
Cross-references
- docs/METHODOLOGY.md §4 plan-execution
- feedback_plan_template_drift.md (stage reality-check)
- feedback_no_tech_debt.md
- /hades:execute-plan slash command handler
