| name | write-plan |
| description | HADES TDD-task-decomposed plan writing: master + stage files pattern,
parallel stage-writer dispatch, watchdog mitigation, two-stage self-review.
Use when operator invokes /hades:write-plan or after design spec is frozen.
|
| license | Proprietary |
| agentskills_version | 1 |
| keywords | ["write-plan","plan-writing","TDD","parallel-writers","watchdog-mitigation","hades"] |
HADES — write-plan skill (plan-writing methodology)
This skill provides HADES's plan-writing methodology, invoked by the
/hades:write-plan slash command. It calls skill_load("superpowers:writing-plans")
then applies project-doctrine overrides.
When to use
- Operator invokes
/hades:write-plan <spec-path>
- After design spec is frozen (6 sections approved + Q&A complete)
Workflow
1. Load writing-plans skill explicitly
skill_load("superpowers:writing-plans")
2. Read + verify spec is frozen
Verify at least 6 design sections approved + Q&A complete per
feedback_spec_hierarchy_and_plan_types.md.
3. Master + stage files pattern
- Master:
design records item-<topic>-master.md (~150-200 lines)
- stage files:
design records item-stage-X-<scope>.md (~700-3000 lines each)
4. Watchdog mitigation (CRITICAL)
Background subagents have 600s stream watchdog. Use incremental Write+Edit per
feedback_plan_writer_watchdog_strategy.md:
- Initial Write with file header + Task 1 only
- Sequential Edits per remaining task
- Final Edit appends verification checklist
5. Two-stage self-review (mandatory for ≥4 parallel writers)
stage — Mechanical greps (placeholder/attribution/uniqueness/module-path/invariant coverage).
stage — Code-reviewer subagent dispatch (MANDATORY; catches cross-stage signature drift).
Cross-references
- docs/METHODOLOGY.md §3 plan-writing
- feedback_methodology_and_conventions.md §3 + §13
- feedback_plan_writer_watchdog_strategy.md
- /hades:write-plan slash command handler
