plan-arc

name: plan-arc description: Orchestrator that generates a complete self-contained arc bundle for an athlete — training plan, nutrition plan, vendored style guides, agent operating instructions, and bundle entry point. Composes plan-training-arc + plan-nutrition-arc + manual writes for README/CLAUDE.md. Trigger when starting a new arc from scratch, or when restructuring a major arc that requires regenerating all bundle components. argument-hint: [athlete-name] [arc-slug] allowed-tools: Read Write Edit Glob Grep Bash SlashCommand

Plan Arc

Produces a complete athletes/<name>/<arc-slug>/ bundle — the self-contained directory a cloud agent pulls and operates against.

The output mirrors the worked example at docs/athletes/andy/arc-2026-summer-dunk/.

When to use

• New arc starting from scratch (athlete + goals + duration agreed)
• Major arc rewrite requiring regenerated training + nutrition + bundle docs
• Onboarding a new athlete (arc number 1)

When NOT to use

• Mid-arc tweaks (use plan-training-arc to regenerate xlsx, or edit individual files directly)
• Adding a new training style to existing arc (vendor it directly into <bundle>/styles/)
• Updating the live current-week.md (that's an execution skill, not a planning skill)

Required inputs

If any are missing, ask the user before proceeding. Don't fabricate.

1. Athlete name + arc slug (e.g., andy, arc-2026-summer-dunk)
2. Athlete profile at athletes/<name>/profile.md — must exist
3. Arc definition:
   • Purpose (1–2 sentences)
   • 2–3 goals (binary, testable, with deadline)
   • Start date + duration
   • Block sequence (count + names + emphasis)
4. Training style choices:
   • Primary style (e.g., vertical-jump)
   • Secondary style (optional, e.g., dylan-shannon)
   • Both must exist as docs/content/training-styles/<style>/guide.md
5. Nutrition style choice:
   • Default: renaissance-diet
   • Must exist as docs/content/nutrition-styles/<style>/guide.md
6. Cross-arc nutrition OS at athletes/<name>/nutrition.md — must exist (defines menu + supply)
7. Hard constraints — injuries, cut/no-cut blocks, equipment limits

Bundle layout (target)

athletes/<name>/<arc-slug>/
├── README.md                    ← bundle entry point + how to pull into agent repo
├── CLAUDE.md                    ← agent operating instructions
├── profile.md                   ← athlete snapshot at arc start
├── training/
│   ├── arc.md                   ← purpose, goals, blocks, testing schedule
│   ├── blocks/<n>-block-<name>.md
│   ├── weeks/<arc>-W<NN>.md     ← all weeks pre-rendered
│   └── active/
│       ├── current-week.md      ← HOT PATH (initialized to Wk 1)
│       └── current-block.md
├── nutrition/
│   └── arc.md                   ← phase per block, bw curve, exception rules
├── styles/
│   ├── <primary-style>-guide.md ← vendored copy
│   └── <secondary-style>-guide.md (if applicable)
└── outputs/
    └── hybrid-athletic-plan-v<n>.xlsx

Workflow

Phase 1: Validate inputs

Check existence of:

• athletes/<name>/profile.md
• athletes/<name>/nutrition.md (cross-arc OS)
• docs/content/training-styles/<primary-style>/guide.md
• docs/content/training-styles/<secondary-style>/guide.md (if specified)
• docs/content/nutrition-styles/<nutrition-style>/guide.md

If any missing, ask the user. Don't proceed.

Confirm the user's stated arc parameters back to them: name, slug, purpose, goals, dates, block sequence, styles. Don't fabricate any of these.

Checkpoint: "Confirm these arc parameters before I start generating?"

Phase 2: Create bundle directory + snapshot profile

mkdir -p athletes/<name>/<arc-slug>/{training/{blocks,weeks,active},nutrition,styles,outputs}

Snapshot the athlete profile:

• Copy athletes/<name>/profile.md to athletes/<name>/<arc-slug>/profile.md
• Update the snapshot's "Last updated" or "Arc start" line to the arc start date
• This is a point-in-time fork — the bundle profile won't track future changes to the cross-arc profile

Phase 3: Generate training arc

Invoke the plan-training-arc skill with the arc inputs. It will produce:

• athletes/<name>/<arc-slug>/outputs/hybrid-athletic-plan-v<n>.xlsx
• And/or markdown files at training/arc.md, blocks/, weeks/, active/

If plan-training-arc only outputs the xlsx, generate the markdown files directly using the same data:

• training/arc.md — purpose, goals, block sequence, testing schedule, constraints, style guide pointers
• training/blocks/<n>-<slug>.md — one per block (programming strategy, weekly split, daily breakdown)
• training/weeks/<arc>-W<NN>.md — pre-render all weeks from the block × week grid
• training/active/current-week.md — initialize to Wk 1
• training/active/current-block.md — initialize to Block 1

Phase 4: Generate nutrition arc

Invoke the plan-nutrition-arc skill with:

• Athlete: <name>
• Arc slug: <arc-slug>
• Nutrition style: <nutrition-style> (default: renaissance-diet)
• It reads training/arc.md (just written in Phase 3), profile, cross-arc OS, style guide
• Produces athletes/<name>/<arc-slug>/nutrition/arc.md

Phase 5: Vendor style guides into the bundle

Copy each style guide:

cp docs/content/training-styles/<primary-style>/guide.md athletes/<name>/<arc-slug>/styles/<primary-style>-guide.md
cp docs/content/training-styles/<secondary-style>/guide.md athletes/<name>/<arc-slug>/styles/<secondary-style>-guide.md  # if applicable

The bundle is self-contained — agents read only from within the bundle. Updates to upstream style guides require a re-vendoring (manual or via a refresh skill).

Note: nutrition style guides are NOT vendored in v0 (the nutrition/arc.md is opinionated enough that the agent doesn't need to read the full nutrition style guide at runtime). Reconsider if/when nutrition-side substitution logic gets richer.

Phase 6: Write bundle README.md

Use the structure from arc-2026-summer-dunk/README.md:

• Title and metadata (start, end, duration, version)
• "What this is" — 3-line summary of the arc
• "Bundle contents" — table mapping each file/dir to "what it is" and "when to read it"
• "How to pull this bundle into your cloud agent repo" — sparse-checkout, rsync, submodule modes
• "Refresh cadence" — when the bundle gets regenerated, how the agent picks it up
• "Storage boundary" — what's NOT in the bundle (Supabase tables for execution data)
• "Versioning" — plan version, source repo commit (TODO if no build script)

Phase 7: Write bundle CLAUDE.md

Use the structure from arc-2026-summer-dunk/CLAUDE.md:

• "Your role" — execution agent for this arc; does NOT design programs
• "Bundle layout" — diagram of the bundle's tree
• "Hot path" — morning prescription routing
• "When the athlete asks 'why?'" — answer from these files in this order
• "Nutrition reading map" — which question routes to which doc
• "Hard constraints (DO NOT violate)" — pull from profile.md injuries + style guide constraints + nutrition arc hard constraints (e.g., "B3 = no cut")
• "When the athlete reports a flare-up" — substitution logic, propose-don't-dictate format
• "Athlete profile cheat sheet" — 5–8 facts to remember
• "Logging behavior" — Supabase write rules
• "Nutrition behavior" — silent in-week, two touches (post-workout bw?, Saturday Costco)
• "When to escalate / propose plan changes" — escalation triggers
• "What's NOT in this bundle" — Supabase, other arcs

This file is hand-tailored per athlete and per arc. It's the most opinionated file in the bundle.

Phase 8: Sanity check

After all files are written:

1. Cross-reference check: every link in CLAUDE.md and README.md resolves to an actual file in the bundle (use Glob / Read to verify).
2. Hard-constraint propagation: any constraint in profile.md is reflected in CLAUDE.md's "Hard constraints" section AND in the relevant style guide's substitution map.
3. Phase coherence: the cut/maintenance pattern in nutrition/arc.md matches the peaking pattern in training/arc.md (e.g., training Wk 13–18 = peak; nutrition Wk 13–18 = maintenance).
4. Self-containment: the bundle has no broken references to files outside athletes/<name>/<arc-slug>/ — except the explicit cross-arc OS pointer in nutrition/arc.md (../../nutrition.md).

Phase 9: Report

Summarize:

• Bundle path and total file count
• Plan version (e.g., v6)
• Arc dates and total weeks
• Block sequence (just names)
• Phase pattern (which blocks cut, maintain, etc.)
• Hard constraints inherited
• TODOs (e.g., "build script for versioning not yet wired")
• Next-action: how to pull this bundle into the agent repo (link to the README's "How to pull" section)

Constraints

• Never overwrite an existing bundle without explicit confirmation. If athletes/<name>/<arc-slug>/ already exists, ask before regenerating.
• Never fabricate athlete data. If profile or arc parameters are unclear, ask. Bundles are durable — wrong data here propagates.
• Always vendor training style guides. Self-containment is the bundle's contract. Don't link to upstream docs from within the bundle.
• Always honor B3 = no cut style hard constraints from training arc. Encode in nutrition/arc.md AND in CLAUDE.md.
• Don't bake build-script versioning into v0. Leave a TODO in README.md's versioning section if no build script exists yet.
• Coordinate with plan-training-arc and plan-nutrition-arc rather than duplicating their logic. This skill is an orchestrator, not a re-implementation.

Outputs

A complete bundle at athletes/<name>/<arc-slug>/:

• 1 README.md, 1 CLAUDE.md, 1 profile.md (snapshot)
• 1 training/arc.md, N block files, N week files, 1 current-week.md, 1 current-block.md
• 1 nutrition/arc.md
• 1–2 vendored style guides
• 1 .xlsx output

Total file count for an 18-week, 3-block arc: ~28 files.