Pick Workflow

Pick Workflow

Authoring-time judge for a skill/agent's execution topology. Sibling to /pick-model: it picks the model+effort for one step; this picks how steps run (linear vs fan-out, sub-agents vs Workflow, the seam) and calls /pick-model per step for the model. Emits a design — never runs the task.

When: authoring or challenging a skill/agent · deciding parallel-vs-linear · sub-agents-vs-Workflow.

Steps

1. Decompose — fill the table FIRST (it IS the analysis; no recommending before it exists)

| Step | Work shape | Parallelizable? | Judgment load | Token weight | Cross-item dep? | |---|---|---|---|---|---|

• Work shape — read / classify / synthesize / transform / decide.
• Parallelizable? — units independent? (per-file yes; "synthesize the whole plan" no.)
• Judgment load — low (mechanical/schema-bound) → high (ordering, trade-offs, reversibility, intent).
• Token weight — where cost sits (usually the step that reads everything).
• Cross-item dep? — compares across units / needs the global graph? Correctness gate: a cross-item step must NOT be sharded. Two gates + worker rules in reference.md.

2. Cut the seam by global-context (NOT by folder/file)

Fan out per-item, judgment-light, independent work → workers return structured facts only (no dispositions/severity/decisions). Keep cross-item reasoning + dispositions + synthesis on the orchestrator. Canonical: gather (fan-out) → reason → plan (orchestrator) → execute (fan-out independent ops).

3. Pick topology + mechanism

Front-door first (Anthropic catalog in reference.md): routing (distinct input kinds → branch before fan-out) · voting / evaluator-optimizer (need confidence/refinement → run-N-vote or generate→critique→refine). Then:

| Choose | When | Gate | |---|---|---| | 🟢 Linear | Below threshold; few units — the default | simplest | | 🔵 Sub-agents (parallel) | fixed small set (2–6) independent tasks, no loop | no resume/budget | | 🟣 Workflow | fan-out over a (variable-size) list; need loop-until-dry, threshold switch, budget cap, schema extraction, verify/voting, resume | opt-in only → never a skill's default; ship a linear fallback |

Every worker needs a delegation contract: objective · output format (schema) · tool/source guidance · boundaries — else duplication/gaps.

4. Threshold, not unconditional fan-out

Fan-out has fixed cost (spawn latency, schema round-trips, orchestrator re-reads). Encode a threshold switch from the target's typical (not worst-case) input. In Workflow: one if (items < N) line.

5. Model+effort per step → call /pick-model

Do not re-derive a model table. Run /pick-model on each step's shape; write its verdict into the design. (Typical: synthesis→Opus, schema extraction→Haiku, careful transforms→Sonnet — but it decides.)

Output

1. Step table (filled). 2. Design — seam + mechanism + threshold + per-step model. 3. Trade-offs — correctness risk avoided · conditionality · honest counter-argument for plain linear. 4. Concrete change — model:/effort: per phase/agent(), spawns or pipeline()/parallel(), threshold, fallback.

Anti-patterns

• ❌ "Parallel sub-agents for exploration" → shard by file → cross-item checks see a slice (the #1 trap).
• ❌ "Cheap model for subtasks" on a judgment step → degraded dispositions/ordering.
• ❌ Default linear+single-model unchecked — wastes tokens AND risks naive over-parallelizing.
• ❌ Workers that judge — they extract facts; dispositions stay on the orchestrator.
• ❌ A skill that requires Workflow — it's opt-in-gated; always ship a linear fallback.
• ❌ Re-deriving a model table — /pick-model's job. ❌ Always-on fan-out — only above threshold.
• ❌ Auto-running this on every SKILL.md edit — it's Opus-shaped; keep it prompt-invoked.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: digital-stoic-org
• Source: digital-stoic-org/agent-skills
• License: MIT

Install and usage instructions live in the source repository linked above.