master-planner

name

master-planner

description

Builds a project master plan as an interconnected web of friction-removing automations aimed at impossible missions. Resolves the skills library from SKILLS_ROOT, ~/Work/personal/skills, or clones https://github.com/p10ns11y/skills when local is missing; pulls a pack into cwd, Orwell-tweaks overlays, and adds ontology for large/complex repos. Use for master plans, skill pull/tweak, project skill packs, ontology, or /master-planner.

Master Planner

Load rule: Formal SoT below. Orwell/ontology depth → references/ only if needed.
CLT: install ../rules/clt-dual-load.mdc with packs (A8 DualLoad pre-filter); ontology/overlays must reduce rediscovery load, not add second SoTs.

// Mission
MasterPlan ≔ friction-removing automation web for an impossible mission (live skill web — not slides)

// Resolve
SKILLS_ROOT → ~/Work/personal/skills → clone https://github.com/p10ns11y/skills
Project ──symlink──► .agents/skills/ | .cursor/skills/
Portable bodies: never fork — overlays only
Project-born skills: keep in-tree; publish via [dual-copy-skill-publish](../dual-copy-skill-publish/SKILL.md)

// Axioms
A1  Plan ⇔ removes friction (tokens | handoffs | rediscovery | unsafe defaults)
A2  Symlink portable skills; overlay tweaks; never dual-edit portable SKILL.md
A3  Ontology only when complex (≥2 of: multi-crate, agent contracts, wrong-file risk, cross-module invariants)
A4  Verify pack + project checks before "done"

Remote: p10ns11y/skills. Companions: ai-optimization · fusion-sage · higher-order-decision-architect · stellar-spacemap · agent-orchestrator

---

When to run

Signal	Action
“Master plan”, “skill pack”, “pull skills”, /master-planner	Full workflow
New repo / cold agent setup	Pull + AGENTS.md + verify
Large complex repo, agents lose the map	Add/refresh ontology
Overlays stale vs shipped architecture	Tweak overlays only

Skip: one-file typo, single bugfix with known path.

---

Workflow (do in order)

- [ ] 1. Orient — cwd, AGENTS.md, existing .agents/skills, mission one-liner
- [ ] 2. Scan library — list skills; map pack to stack (see Pack router)
- [ ] 3. Pull — symlink missing skills; keep in-repo skills that are project-owned
- [ ] 4. Tweak — Orwell + action + diagram + compress overlays (see Tweak law)
- [ ] 5. Ontology — if complex: INDEX + graph (+ GRAPH.md); else skip
- [ ] 6. Wire — AGENTS.md active-skills table + verify block
- [ ] 7. Verify — script + project checks; report gaps

1. Orient

pwd
test -f AGENTS.md && head -80 AGENTS.md
ls -la .agents/skills 2>/dev/null || ls -la .cursor/skills 2>/dev/null

Capture: mission (one sentence) · stack signals (Cargo.toml, install.sh, package.json, …) · already wired skills.

2. Scan library

# Resolves local or clones https://github.com/p10ns11y/skills
SKILLS_ROOT="$(./scripts/resolve-skills-root.sh)"
ls "$SKILLS_ROOT" | grep -vE '^(examples|rules|\.git|LICENSE|README)'

Read library README.md pack table. Prefer starter packs over one-off picks.

3. Pull

# Resolves SKILLS_ROOT (local → remote clone) then symlinks
./scripts/pull-skills.sh --project "$(pwd)" --pack <pack-name>
./scripts/pull-skills.sh --project "$(pwd)" --skills ai-optimization,fusion-sage,...
# force remote even if a stale local path exists:
./scripts/pull-skills.sh --project "$(pwd)" --pack arch-guardian --from-remote

Rules:

• Symlink into .agents/skills/<name> (arch-machine style) or .cursor/skills/ — match repo convention.
• Never edit portable SKILL.md bodies for one project — use .agents/overlays/ or examples/overlays/.
• Keep in-repo skills (not symlinks) when they encode project-only architecture (e.g. eagle TEA).

4. Tweak (overlays)

Load references/orwell-tweak.md. For each pulled skill that needs local coords:

Write	Path
Relevance / never-compress	.agents/overlays/<project>-ai-optimization.md
Fused aggregates	.agents/overlays/<project>-fusion-sage.md
Decision zones	.agents/overlays/<project>-decision-hooks.md
Roadmap scout	.agents/overlays/<project>-stellar-spacemap.md
Master pack map	.agents/overlays/<project>-master-planner.md

Tweak law (non-negotiable):

1. Orwell — short words; cut dead metaphors; prefer active verbs; one idea per line.
2. Action — every section ends in a verb the agent can run (path, command, refuse).
3. Diagram — one mermaid or ASCII arch for the fused loop; drop second prose copy.
4. Compress — tables > paragraphs; drop legacy unless still live; no token waste.
5. Specific — real paths and verify commands for this cwd.

5. Ontology (large / complex only)

Create when ≥2 of: multi-crate/control-plane, agent-facing contracts, easy wrong-file edits, cross-module invariants.

Assets (template: references/ontology-template.md):

File	Role
.agents/ontology/INDEX.md	Intent → subgraph router
.agents/ontology/<project>.graph.yaml	Nodes + edges + source_refs
.agents/ontology/GRAPH.md	Optional mermaid viz

Point agents: load subgraph by intent; never dump full graph.

6. Wire AGENTS.md

Update Active skills table: skill · path · when. Point overlays. Keep verify block exact.

7. Verify

./scripts/verify-pack.sh --project "$(pwd)"
# then project checks from AGENTS.md (lint, tests, --validate, …)

Done when:

• Pack symlinks resolve
• Overlays match live architecture (no dead entrypoints)
• Ontology (if any) subgraphs route without contradiction
• AGENTS.md lists active skills + verify commands
• Report: pack · mission · friction removed · next automation

---

Pack router

Stack signals	Pack	Core skills
install.sh + profiles + maintenance + archy/TUI	arch-guardian	master-planner, eagle-*, ai-optimization, fusion-sage, HODA, stellar-spacemap, verification-cockpit, session-unit-order, agent-orchestrator, control-graph, git-worktrees
~/.config/shell / path.contract	shell-verify	shell-kernel-ontology, verification-cockpit, stellar-spacemap, ai-optimization, fusion-sage
Next.js / React app	web-app	ai-optimization, fusion-sage, react-client-expert, semantic-markup-css, fix-dependency-security
Tauri + agent loops	agentic-desktop	finder-reactor, tauri-agentic, agent-orchestrator, git-worktrees, control-graph
Multi-agent delivery only	multi-agent	agent-orchestrator, control-graph, git-worktrees, concurrent-cli-agents, split-to-prs
Strategy / backlog docs	strategy	HODA, stellar-spacemap, fusion-sage, master-planner, odysseus-navigator

Custom: compose from catalog; document in overlay.

---

Master plan output (when user wants the plan, not only the pull)

## Master plan — <project>

**Mission:** <one sentence — the impossible made routine>

**Friction web:**
| Automation | Removes | Skill / path | Verify |
|------------|---------|--------------|--------|

**Arch:**
\`\`\`mermaid
flowchart LR
  ...
\`\`\`

**Refuse / build:** …
**Next wave:** …

Pair roadmap depth with stellar-spacemap. Pair material bets with higher-order-decision-architect.

---

Anti-patterns

¬	Do
prose roadmap, no automation edges	friction web table + skills
fork portable SKILL.md per project	overlays + symlink
pull whole library “just in case”	pack router
ontology = AGENTS.md copy	addressable IDs + subgraphs
dead shims as primary	update or delete overlays
funeral close	name acceleration automation

Refs: orwell-tweak · ontology-template · arch-machine-pack · master-planner.mdc