Meta-Planner

Plan-mode-only skill. Produces an implementation plan informed by helix's accumulated project insights.

Environment

HELIX="$(cat .helix/plugin_root)"

Phases: RECALL → EXPLORE → PLAN → SYNTHESIZE → EXIT

1. RECALL

python3 "$HELIX/lib/injection.py" strategic-recall "$ARGUMENTS"

Parse JSON. Use summary for triage, synthesize insights into blocks:

1. CONSTRAINTS — proven insights (_effectiveness >= 0.70): decomposition rules, verification needs, sequencing.
2. RISK_AREAS — risky insights (_effectiveness < 0.40) or derived/failure tags: flag for extra verification, smaller tasks.
3. EXPLORATION_TARGETS — areas referenced by insights that expand scope beyond the naive objective.
4. GRAPH_DISCOVERED — _hop: 1 insights (graph-adjacent, not direct match). Treat as exploration targets.

Triage signals: coverage_ratio > 0.3 = well-mapped, trust constraints. < 0.1 = uncharted, expand exploration. graph_expanded_count > 0 = graph surfacing related context.

If recall returns empty -- proceed without constraints; first sessions have no memory.

2. EXPLORE

Map areas relevant to both objective and insight-identified targets.

1. git ls-files | head -80 -- identify 3-6 natural partitions.
2. Select partitions: union of (a) obviously relevant to objective and (b) areas flagged by RECALL insights.
3. Spawn explorer swarm: subagent_type="helix:helix-explorer", model=sonnet, max_turns=30. All in ONE message. Prompt: SCOPE: {partition}\nFOCUS: {focus}\nOBJECTIVE: $ARGUMENTS.
4. Merge findings by file path. Proceed with successful explorers on error.

Obvious scope (single module, clear file set): skip swarm, use Glob/Grep/Read directly.

3. PLAN

Spawn planner: subagent_type="helix:helix-planner", max_turns=500. Prompt: OBJECTIVE: $ARGUMENTS\nEXPLORATION: {merged_findings_json}\nCONSTRAINTS: {constraints_from_recall}\nRISK_AREAS: {risk_areas_from_recall}. Omit empty blocks. Parse PLAN_SPEC JSON array.

If decomposition raises questions -- use AskUserQuestion to resolve before synthesis.

4. SYNTHESIZE

Write the plan file (path from system context):

# {Objective summary}

## Context
Why this change is needed — the problem, what prompted it, intended outcome.

## Insights Applied
Relevant helix insights and how each shaped the plan:
- [eff%] insight content → influenced {which decision}

## Key Files
Files identified by exploration, grouped by concern:
- {area}: `file1.py`, `file2.py` — {what they do, why they matter}

## Implementation Plan

### 1. {slug} (seq)
{description}
- **Files:** relevant_files
- **Depends on:** blocked_by (or "none — parallel")
- **Verify:** command

### 2. {slug} (seq)
...

## Verification
How to test the complete change end-to-end.

Quality bar: A developer reading the plan should know exactly what changes, in what order, verified how — without re-exploring the codebase.

5. EXIT

Call ExitPlanMode to present the plan for user approval.