Claude-skill-registry edird-phase-planning
Apply when doing planning for long-running tasks in sessions on top level
install
source · Clone the upstream repo
git clone https://github.com/majiayu000/claude-skill-registry
Claude Code · Install into ~/.claude/skills/
T=$(mktemp -d) && git clone --depth=1 https://github.com/majiayu000/claude-skill-registry "$T" && mkdir -p ~/.claude/skills && cp -r "$T/skills/data/edird-phase-planning" ~/.claude/skills/majiayu000-claude-skill-registry-edird-phase-planning && rm -rf "$T"
manifest:
skills/data/edird-phase-planning/SKILL.mdsource content
EDIRD Phase Planning
When to Invoke
or/build
workflows/solve- [PLAN] - creating high-level plans to achieve goals
- Planning for long agentic runs for features, fixes, or research
NOT for document writing - Use dedicated workflows instead:
for SPEC documents/write-spec
for IMPL documents/write-impl-plan
for TEST documents/write-test-plan
for TASKS documents/write-tasks-plan
Quick Reference
Phases: EXPLORE → DESIGN → IMPLEMENT → REFINE → DELIVER
Workflow types: BUILD (code output) | SOLVE (knowledge/decision output)
Assessment: COMPLEXITY-LOW/MEDIUM/HIGH | PROBLEM-TYPE (RESEARCH/ANALYSIS/EVALUATION/WRITING/DECISION)
Phase Gates
EXPLORE → DESIGN
- Problem or goal clearly understood
- Workflow type determined (BUILD or SOLVE)
- Assessment complete (BUILD: COMPLEXITY | SOLVE: PROBLEM-TYPE)
- Scope boundaries defined
- No blocking unknowns requiring [ACTOR] input
DESIGN → IMPLEMENT
- Approach documented (outline, spec, or plan)
- Risky parts proven via POC (if COMPLEXITY-MEDIUM or higher)
- No open questions requiring [ACTOR] decision
- For BUILD: SPEC, IMPL, TEST documents created
- For BUILD: TASKS document created via [PARTITION]
- For SOLVE: Structure/criteria validated
IMPLEMENT → REFINE
- Core work complete (code written / document drafted)
- For BUILD: Tests pass
- For BUILD: No TODO/FIXME left unaddressed
- For SOLVE: All sections drafted
- Progress committed/saved
REFINE → DELIVER
- Self-review complete
- Verification against spec/rules passed
- For BUILD COMPLEXITY-MEDIUM+: Critique and reconcile complete
- For SOLVE: Claims verified, arguments strengthened
- All found issues fixed
Workflow Examples
BUILD (COMPLEXITY-HIGH)
[EXPLORE] → [RESEARCH] → [ANALYZE] → [ASSESS] → [SCOPE] → Gate [DESIGN] → [PLAN] → [WRITE-SPEC] → [WRITE-IMPL-PLAN] → [PROVE] → [PARTITION] → Gate [IMPLEMENT] → [IMPLEMENT] → [TEST] → [FIX] → [COMMIT] → Gate (loop until green) [REFINE] → [REVIEW] → [VERIFY] → [CRITIQUE] → [RECONCILE] → Gate [DELIVER] → [VALIDATE] → [MERGE] → [CLOSE] → [ARCHIVE]
SOLVE (EVALUATION)
[EXPLORE] → [RESEARCH] → [ANALYZE] → [ASSESS] → EVALUATION → Gate [DESIGN] → [FRAME] → [OUTLINE] criteria → [DEFINE] framework → Gate [IMPLEMENT] → [RESEARCH] options → [EVALUATE] → [SYNTHESIZE] → Gate [REFINE] → [CRITIQUE] → [VERIFY] claims → [IMPROVE] → Gate [DELIVER] → [CONCLUDE] → [RECOMMEND] → [VALIDATE] → [ARCHIVE]
Note: COMPLEXITY-LOW skips [PROVE], [CRITIQUE], [RECONCILE].
Phase Plan Requirements
Plans created via [PLAN] must define:
- Objectives - What success looks like
- Strategy - How to achieve objectives
- Deliverables - Concrete outputs with checkboxes
- Transitions - When to move to next phase
Planning Horizon:
[EXPLORE] ← Plan now [DESIGN] ← Plan now [IMPLEMENT] ← TBD (after DESIGN gate) [REFINE] ← TBD (after IMPLEMENT gate) [DELIVER] ← Plan now (shipping tasks from NOTES)
How to Plan Well
Goal Decomposition
- Start with outcome - What does "done" look like?
- Identify dependencies - What must complete before what?
- Find parallel opportunities - What can run concurrently?
- Size steps for testability - Each step should be verifiable
Scope Calibration
- Too big: Step takes >30min AWT or touches >3 files → decompose further
- Too small: Step takes <2min AWT → combine with adjacent step
- Right size: Verifiable outcome, clear done criteria, single responsibility
Dependency Mapping
Ask for each step:
- What inputs do I need? (determines predecessors)
- What outputs do I produce? (determines successors)
- Can this run while something else runs? (candidate for Concurrent block)
Common Planning Mistakes
- Vague objectives: "Make it work" → Define specific success criteria
- Missing dependencies: Steps that assume prior work → Explicit
← Px-Sy - Over-sequencing: All steps linear when some could parallelize
- Under-estimating: No AWT in Strategy → Add time budget
- Skipping verification: No test step after implement → Add [TEST] or [VERIFY]
Next Action Logic
- Check phase gate → Pass? → Next phase, first verb
- Gate fails? → Execute verb that addresses unchecked item
- Verb outcome: -OK → next verb | -FAIL → handle | -SKIP → next verb
- No more verbs? → Re-evaluate gate
- [DELIVER] done? → [CLOSE] and [ARCHIVE] if session-based
Common failure handlers:
- -FAIL on [RESEARCH], [ASSESS], [PLAN] → [CONSULT] or more [RESEARCH]
- -FAIL on [TEST], [VERIFY] → [FIX] → retry
Effort Allocation
Time Units
- AWT (Agentic Work Time) - Agent processing time, excludes user wait
- HHW (Human-Hour Work) - Human equivalent effort for task sizing
Phase Budgets by Complexity
| Phase | LOW | MEDIUM | HIGH |
|---|---|---|---|
| EXPLORE | 5min | 15min | 30min |
| DESIGN | 5min | 30min | 60min |
| IMPLEMENT | varies | varies | varies |
| REFINE | 5min | 15min | 30min |
| DELIVER | 2min | 5min | 10min |
Values are AWT guidelines, not hard limits.
Diminishing Returns
- Phase takes 2x budget without gate progress → [CONSULT]
- Same step retried 3x without improvement → [CONSULT]
- Research yields no new information after 3 sources → move on
Retry Limits
- COMPLEXITY-LOW: Infinite retries (until user stops)
- COMPLEXITY-MEDIUM/HIGH: Max 5 attempts per phase, then [CONSULT]
Mandatory Gate Output
Before proceeding to next phase, output:
## Gate: [CURRENT_PHASE] → [NEXT_PHASE] **Complexity**: [LOW/MEDIUM/HIGH] | **Artifacts**: [list created docs] - [x] Item - Evidence: [specific evidence] - [ ] Item - BLOCKED: [what's missing] **Gate status**: PASS | FAIL