Pick 3 Chain

Input: $ARGUMENTS

Core Principles

1. Chains are workflows, not lists. The 3 skills must form a connected sequence where each skill’s output feeds the next skill’s input. A list of 3 unrelated skills is not a chain — it’s a menu.

2. Three is a complete arc. A 3-skill chain should form a complete thinking arc: setup -> core operation -> validation (or input -> transform -> output). Each position in the chain has a role.

3. Follow invocation links first. The skill graph already encodes workflows through invokes/invoked_by. Start by tracing these existing links before constructing novel chains.

4. Chains run in one direction. The execution order must be unambiguous. Step 1 produces output that Step 2 requires. Step 2 produces output that Step 3 requires. No loops, no optional branches.

5. The seed determines the chain. The user provides a starting skill. The chain builds OUTWARD from that seed — either forward (what the seed invokes) or backward (what invokes the seed) or both.

Phase 1: Seed Analysis

[A] SEED: /[skill from $ARGUMENTS]
[B] SEED_FUNCTION: [what does this skill do?]
[C] SEED_OUTPUTS: [what does it produce?]
[D] SEED_INVOKES: [skills it calls — forward chain candidates]
[E] SEED_INVOKED_BY: [skills that call it — backward chain candidates]
[F] SEED_CATEGORY: [category from skills.json]

Read the seed skill’s SKILL.md to understand its inputs and outputs.

[G] SEED_INPUTS: [what does the seed skill need as input?]
[H] SEED_GAPS: [what does the seed NOT do that the user might need?]

Phase 2: Chain Construction

Three chain strategies, tried in order of preference:

Strategy 1: Forward Chain (seed is Step 1)

[I] FORWARD:
    Step 1: SEED → produces [output]
    Step 2: Which invoked skill best uses that output? → select
    Step 3: What does Step 2 produce? → Which skill best uses THAT? → select

    Chain: SEED → [step 2] → [step 3]

Strategy 2: Backward Chain (seed is Step 3)

[J] BACKWARD:
    Step 3: SEED ← needs [input]
    Step 2: Which invoking skill produces that input? → select
    Step 1: What does Step 2 need? → Which skill produces THAT? → select

    Chain: [step 1] → [step 2] → SEED

Strategy 3: Center Chain (seed is Step 2)

[K] CENTER:
    Step 2: SEED ← needs [input], produces [output]
    Step 1: Which skill produces what SEED needs? → select
    Step 3: Which skill uses what SEED produces? → select

    Chain: [step 1] → SEED → [step 3]

Strategy Selection

[L] STRATEGY_SELECTION:
    IF seed has strong forward invocations → FORWARD
    IF seed has strong backward invocations → BACKWARD
    IF seed has both → CENTER
    IF seed has neither → construct novel chain based on function/category

    SELECTED: [strategy]

Phase 3: Chain Validation

[M] VALIDATION:

Step 1: Does Step 1's output match Step 2's input? [Y/N — describe the connection]
Step 2: Does Step 2's output match Step 3's input? [Y/N — describe the connection]
Step 3: Is there a clear purpose for this chain? [state it in one sentence]
Step 4: Could the chain run in reverse and still make sense? [Y/N — if Y, the chain logic may be weak]
Step 5: Does each step add something the previous step couldn't do? [Y/N for each]

Phase 4: Output

ALGORITHM: CHAIN
SEED: /[seed skill]
CHAIN LENGTH: 3
STRATEGY: [forward / backward / center]

CHAIN:
  Step 1: /[id] — [title]
     Role: [what it does in this chain]
     Produces: [output that feeds Step 2]
     ↓ feeds into
  Step 2: /[id] — [title]
     Role: [what it does in this chain]
     Receives: [from Step 1] | Produces: [output that feeds Step 3]
     ↓ feeds into
  Step 3: /[id] — [title]
     Role: [what it does in this chain]
     Receives: [from Step 2] | Produces: [final output]

CHAIN PURPOSE: [1-sentence: what does running this chain accomplish?]
CHAIN ARC: [setup → core → validation / input → transform → output / etc.]

ALTERNATIVE CHAINS:
  Alt 1: /[id] → /[id] → /[id] — [1-line what this chain does differently]
  Alt 2: /[id] → /[id] → /[id] — [1-line]

Failure Modes

Depth Scaling

Default: 2x. These are floors.

Pre-Completion Checklist

• Seed skill read and its inputs/outputs identified
• Chain strategy selected with rationale
• Exactly 3 skills in the chain (including seed)
• Each step’s output feeds the next step’s input
• Chain purpose stated in one sentence
• Each step adds a unique capability
• At least 1 alternative chain provided

Integration

• Shortcut for: /pick 3 chain $ARGUMENTS
• Use when: You have a skill and want to know what comes before/after it
• Routes to: The 3 chain skills in sequence; /to for longer workflow design
• Related: /p10complement (broader pairing), /de (dependency extraction)
• Differs from /p10complement: complement finds 10 partners; chain finds 3 in sequence
• Differs from /de: de extracts dependencies from a project; chain builds a skill workflow