extract-skill-extractor

name	extract - Skill Extractor
description	Given any prompt, decompose it into implicit sub-tasks and extract every potentially useful skill, grouped by role and prioritized with invocation-ready suggestions.
output	{"format":"hybrid"}

Extract - Skill Extractor

Input: $ARGUMENTS

---

Core Principles

1. Prompts are layered. A single sentence can contain a surface request, an implicit prerequisite, a validation need, and an emotional undercurrent. Extract skills for all layers, not just the obvious one.

2. Role determines value. A skill can serve as primary (does the core work), preparatory (should run first), supporting (enhances output), validation (checks output), or recovery (useful if primary fails). The same skill in different roles has different priority.

3. Sequence is load-bearing. Running /aex before /dcp changes the decision. Running /gu before /want changes the goal. Skill ordering is analytical work, not decoration.

4. Implicit beats explicit. The user who says "help me write a proposal" probably also needs /gu (clarify the goal), /aex (check assumptions about the audience), and /pv (validate the draft). The unexpressed needs are where extraction earns its keep.

5. Interaction effects exist. Some skill pairs amplify each other (/aex + /ht). Some conflict (/ma diverges, /dcp converges — sequence matters). Flag these.

---

Phase 1: Prompt Decomposition

Break the input into every distinct sub-need. Go beyond the literal request.

[E1] SUB-NEED: [label] — [one-line description]
[E2] SUB-NEED: [label] — [one-line description]
...

Apply all decomposition lenses that yield results:

Lens	Question
Surface	What is the user literally asking for?
Prerequisite	What must be true or known before the surface task can succeed?
Assumption	What is the user assuming that might be wrong?
Validation	How would the user know the output is good?
Emotion	Is frustration, uncertainty, or overwhelm present?
Scope	Is the user scoping too narrowly or too broadly?
Sequence	Does this need to happen in a particular order?
Downstream	What will the user do with the result?

---

Phase 2: Skill Mapping

For each sub-need, identify candidate skills and assign a role:

Role	Meaning	Priority Weight
Primary	Directly accomplishes the sub-need	Highest
Preparatory	Should run before the primary to improve its input	High
Supporting	Enhances or deepens the primary's output	Medium
Validation	Checks the primary's output for quality or correctness	Medium
Recovery	Useful if the primary fails or produces weak results	Low

[E-N] SKILL: /skill-id → sub-need [E-ref] as [role] — [one sentence why]

---

Phase 3: Prune and Prioritize

Remove weak matches. A match is weak if:

• The skill only tangentially relates to the sub-need
• A stronger skill already covers the same ground in the same role
• The skill requires context the user clearly does not have

Rank surviving skills by: role_weight x sub-need_importance x fit_quality

Group into tiers:

• ESSENTIAL — Skip these and the output quality drops significantly
• RECOMMENDED — Meaningful improvement, worth the time
• OPTIONAL — Useful if the user wants thoroughness

---

Phase 4: Interaction Analysis

Identify skill pairs or sequences where order matters or effects compound.

[E-N] INTERACTION: [type] — /skill-a + /skill-b: [what happens and why order matters]

Type	Description
Amplifying	Running both produces more than the sum
Sequenced	One must run first or the second's input is degraded
Conflicting	Running both without awareness creates tension
Redundant	High overlap; pick one unless depth warrants both

---

Phase 5: Output

PROMPT: [original input, quoted]

ESSENTIAL SKILLS (ordered by execution sequence):
1. /skill-id — ROLE: [role] — WHY: [one line] — INVOKE: /skill-id [suggested args]
2. ...

RECOMMENDED SKILLS:
3. /skill-id — ROLE: [role] — WHY: [one line] — INVOKE: /skill-id [suggested args]
...

OPTIONAL SKILLS:
- /skill-id — ROLE: [role] — WHY: [one line]

INTERACTION EFFECTS:
- [skill-a] → [skill-b]: [effect and recommended order]

SUGGESTED SEQUENCE:
[skill-a] → [skill-b] → [skill-c]
(brief rationale for ordering)

---

Failure Modes

Failure	Signal	Fix
Surface-only extraction	Only the literal request mapped	Apply all 8 decomposition lenses
Skill-name anchoring	Picked skills whose names sound right	Verify by checking what the skill actually does
Over-extraction	20 skills for a simple prompt	Prune ruthlessly; simple prompt = short output
Role collapse	Everything assigned "primary"	Force at least 2 distinct roles at 2x+ depth
Flat ordering	Skills listed alphabetically	Phase 4 interaction analysis must drive sequence
Missing prerequisites	Recommended skill needs input user hasn't produced	Check each skill's input requirements
Ignoring emotional layer	Only analytical skills extracted	Emotion lens catches frustration/uncertainty

---

Depth Scaling

Depth	Typical Sub-Needs	Typical Skills	Typical Interactions	Typical Roles
1x	1-3	2-5	0-1	1
2x	3-5	6-10	1-3	2-3
4x	4-7	10-16	3-5	3-4
8x	6-10	16-25	5-8	4-5

Default: 2x. These are guidelines, not floors. A simple prompt with 2 genuine sub-needs and 3 perfect skills is better than padding to hit a number.

---

Pre-Completion Checklist

• Every decomposition lens applied (not just surface)
• Each extracted skill has an explicit role assignment
• At least one non-obvious skill identified
• Interaction effects between top skills analyzed
• Suggested sequence reflects dependency and amplification order
• Invocation examples include concrete arguments from user's prompt
• No skill included without a specific "why" tied to a sub-need
• Over-extraction checked: every ESSENTIAL skill genuinely earns its tier

---

Integration

• Use from: /meta, /handle, /fonss
• Routes to: any extracted skill via invocation suggestions
• Differs from /wsib: wsib picks ONE skill; extract maps the full landscape
• Differs from /given: given ranks by ROI for a goal; extract maps by coverage of sub-needs
• Differs from /fonss: fonss orders next skills; extract finds all relevant skills