shallow · v2 · 2026-10-01 · sha256 854ded0e0175db3d
shallow v2A
Immutable. This exact content never changes and is served at /api/v1/blob/854ded0e0175db3d.
No tool to install first
claude plugin marketplace add https://markdownregistry.com/mp/a/art_yfhtko457dq4vw77/854ded0e0175db3dea746588113f5667691d15a0417197be2c84fe34d0321c82.json
Then /plugin install shallow@markdownregistry-yfhtko457dq4vw77 in Claude Code. It installs this exact version, the one graded A here, pinned to the commit it was graded at, and leaves it there.
---
name: shallow
codename: SHALLOW
internal: Proportional Depth Protocol v2.0
version: 2.0
tier: cognition
trigger: low-stakes task, reversible decision, "keep it simple", "don't overthink", quick answer needed
description: Prevents overthinking by matching cognitive investment to consequence - the intentional opposite of DEEP-THINK. Now with mathematical depth metrics, depth budget calculus, proportionality , and verification gates.
author: Kshitijpalsinghtomar
tags: [proportionality, efficiency, overthinking, quick-answer, depth-calibration, metrics, recursion, verification]
artifacts:
- depth-assessment
- budget-allocation
- shallow-delivery-verdict
- depth-budget-calculus
- activation-heatmap
- phase-activation
composable_with:
- deep-think
- threshold
- conductor
- boundary-detector
thinking_parameters:
require_calculus: true
require_proportionality_: true
recursion_depth: 1
---
# SHALLOW v2.0 — Proportional Depth Protocol (Premium)
> **Mathematical Compliance**: This skill implements the DEPTH-MATHEMATICS specification. It computes and outputs ADS phase activation, CEI components, SI components, cognitive traces (depth budget calculus, proportionality ), and verifies Gates V1, V6. It formalizes proportional depth as a budget calculus.
DEEP-THINK is essential. But not every task deserves it. Sometimes the deepest answer is a short one.
The failure mode of depth-skills is **depth bias**: the model reaches for sophisticated analysis on problems that don't warrant it. A CSS color choice gets 6-step reasoning. A routine email gets threat-modeled. This skill is the brake.
**New in v2.0:** Mathematical depth metrics, depth budget calculus, proportionality , recursive self-audit, verification gates.
---
## The Failure Mode You Must Recognize
You are about to:
- Generate six artifacts for a question that could be answered in one sentence
- Add caveats to a decision that is easily reversed
- Apply THRESHOLD analysis to a temporary workaround
- Treat a first-pass response as if it were a permanent architecture decision
If the cost of being wrong is low and the fix is easy — this is a SHALLOW task.
**Cargo cult shallowness** gives short answers without justifying why depth isn't needed. This skill rejects it.
---
## The Protocol
### 0 — PARAMETERIZE: Read Thinking Parameters
Read `thinking_parameters`:
- `require_calculus` (default true) — compute depth budget calculus
- `require_proportionality_` (default true) — issue proportionality
- `recursion_depth` (default 1) — recursive self-audit rounds
---
### 1 — DEPTH BUDGET CALCULUS (if `require_calculus` = true)
**NEW IN v2.0** — Formal calculus for depth appropriateness.
Evaluate the task against depth criteria:
```
DEPTH BUDGET CALCULUS
────────────────────────────────────────
Reversibility: [Trivial / Moderate / Hard / Permanent]
Weight: [1.0 / 0.5 / 0.2 / 0.0] → Score: [value]
Consequence: [Trivial / Moderate / Significant / Critical]
Weight: [1.0 / 0.5 / 0.2 / 0.0] → Score: [value]
Complexity: [Simple / Moderate / Complex / Intricate]
Weight: [1.0 / 0.5 / 0.2 / 0.0] → Score: [value]
Pattern match: [Routine / Variation / Novel / Unprecedented]
Weight: [1.0 / 0.5 / 0.2 / 0.0] → Score: [value]
Stakeholders: [One / Few / Many / All-hands]
Weight: [1.0 / 0.5 / 0.2 / 0.0] → Score: [value]
Urgency: [Can wait / Within hour / Within minute / Now]
Weight: [1.0 / 0.5 / 0.2 / 0.0] → Score: [value]
────────────────────────────────────────
TOTAL SHALLOW SCORE: Σ Scores / 6 = [0.0-1.0]
(1.0 = fully shallow, 0.0 = fully deep)
Depth Budget:
SHALLOW SCORE ≥ 0.67 → MINIMAL (no artifacts, one sentence)
0.33 ≤ SHALLOW SCORE < 0.67 → LIGHT (one artifact, one assumption, one alternative)
SHALLOW SCORE < 0.33 → MODERATE (restatement + approach, skip challenge)
────────────────────────────────────────
```
**Artifact:** `depth_budget_calculus` — formal justification for depth level.
**CEI Component:** Depth Budget Calculus (weight 3.0)
---
### 2 — ALLOCATE COGNITIVE BUDGET
Based on the calculus, allocate:
```
BUDGET ALLOCATION
────────────────────────────────────────
Shallow Score: [0.XX]
Depth Budget: [MINIMAL / LIGHT / MODERATE]
MINIMAL (Score ≥ 0.67):
- No artifacts required
- One sentence answer
- Skip caveats
- Skip alternatives
- Skip "it depends"
- ADS Target: 0.15
LIGHT (0.33 ≤ Score < 0.67):
- One artifact: answer with ONE assumption stated
- One alternative mentioned in 1 sentence
- One caveat if genuinely important
- ADS Target: 0.30
MODERATE (Score < 0.33):
- Two artifacts: restatement + chosen approach
- Skip challenge phase
- Minimal boundary check
- ADS Target: 0.45
────────────────────────────────────────
```
**Artifact:** Budget allocation with ADS targets.
**CEI Component:** Budget Allocation (weight 2.0)
---
### 3 — DELIVER WITHIN BUDGET
Execute the answer within allocated depth:
```
SHALLOW DELIVERY
────────────────────────────────────────
Budget level: [MINIMAL / LIGHT / MODERATE]
Answer: [deliver within budget constraints]
If MINIMAL:
Answer only. No preamble. No caveats.
If LIGHT:
State ONE assumption. Mention ONE alternative.
One sentence each, inline.
If MODERATE:
Restate in one sentence.
Give the answer.
One-line boundary: "works for [common case], not [edge case]."
────────────────────────────────────────
```
**Artifact:** The shallow delivery.
**CEI Component:** Shallow Delivery (weight 1.0)
---
### 4 — PROPORTIONALITY (if `require_proportionality_` = true)
**NEW IN v2.0** — Formal of proportional depth.
```
PROPORTIONALITY
────────────────────────────────────────
ID: pc_<timestamp>_<hash>
Task: [description]
Depth Budget Calculus:
Reversibility: [level] → [score]
Consequence: [level] → [score]
Complexity: [level] → [score]
Pattern Match: [level] → [score]
Stakeholders: [level] → [score]
Urgency: [level] → [score]
Total Shallow Score: [0.XX]
Depth Budget: [MINIMAL / LIGHT / MODERATE]
ADS Target: [0.XX]
Actual ADS: [0.XX] (from CONDUCTOR)
Proportionality: [PROPORTIONAL / OVER-DEEP / UNDER-DEEP]
Proportional: Actual ADS within ±0.1 of target
Over-Deep: Actual ADS > target + 0.1 (wasted cognitive effort)
Under-Deep: Actual ADS < target - 0.1 (insufficient analysis)
Status: [PROPORTIONAL / OVER-DEEP / UNDER-DEEP]
Valid Until: [date or condition]
────────────────────────────────────────
```
**Artifact:** `proportionality_` — formal verification of proportional depth.
**CEI Component:** Proportionality (weight 2.0)
---
### 5 — METRICS: Compute and Output Depth Metrics
#### 5.1 Phase Activation (for ADS)
```json
{
"phase": 1,
"skill": "shallow",
"artifact_completeness": 0.XX,
"gate_pass_rate": 0.XX,
"external_validity": 0.XX,
"human_eval": 0.XX,
"phase_activation": 0.XX
}
```
#### 5.2 CEI Components
```json
{
"cei_components": {
"depth_budget_calculus": 3.0,
"budget_allocation": 2.0,
"shallow_delivery": 1.0,
"proportionality_": 2.0,
"total_complexity_weight": 8.0,
"output_tokens": N,
"wall_time_seconds": T,
"cei": 0.XX
}
}
```
#### 5.3 SI Components
```json
{
"si_components": {
"c2_c3_count": 0,
"contrarian_viable": false,
"fatal_attacks": 0,
"boundary_violations": 0,
"assumption_reversals": 0,
"proportionality_violations": 0,
"si_total": 0
}
}
```
#### 5.4 Cognitive Traces
**Depth Budget Calculus:**
```json
{
"type": "depth_budget_calculus",
"dimensions": [
{"dimension": "Reversibility", "level": "...", "score": 0.XX},
{"dimension": "Consequence", "level": "...", "score": 0.XX},
{"dimension": "Complexity", "level": "...", "score": 0.XX},
{"dimension": "Pattern Match", "level": "...", "score": 0.XX},
{"dimension": "Stakeholders", "level": "...", "score": 0.XX},
{"dimension": "Urgency", "level": "...", "score": 0.XX}
],
"total_shallow_score": 0.XX,
"depth_budget": "MINIMAL/LIGHT/MODERATE",
"ads_target": 0.XX
}
```
**Proportionality :** (as defined in Step 4)
#### 5.5 Depth Contribution
```json
{
"skill": "shallow",
"version": "2.0",
"metrics": {"ads_contribution": 0.XX, "cei": 0.XX, "si": 0, "ec": 0.XX},
"gates_verified": {"V1": true, "V6": true},
"limitations": ["..."]
}
```
---
### 6 — GATES: Verify Before Delivery
**MANDATORY** — Verify these gates PASS:
| Gate | Check | Pass Condition |
|------|-------|----------------|
| **V1** | Assumption Coverage | If MODERATE budget: every claim traces to assumption |
| **V6** | Recursive Stability | If recursion: RSM > 0.95 ∧ SI not decreasing |
**If any gate FAILS:** Return to relevant step and fix. Do not deliver.
---
### 7 — RECURSIVE SELF-AUDIT (if `recursion_depth` > 1)
If `recursion_depth` > 1, apply **this entire protocol** to your own output from Steps 1-4.
For each recursion level d = 2 to `recursion_depth`:
1. Treat your previous output as the "answer under review"
2. Run Steps 1-4 on it
3. Compute **RSM** (Recursive Stability Metric):
```
RSM = 1 - Semantic_Distance(output_d, output_{d-1})
```
4. **Convergence Check:**
- If RSM > 0.95 AND SI_d ≥ SI_{d-1} for 2 consecutive depths → **CONVERGED**, stop
- If RSM < 0.80 OR SI_d < SI_{d-1} for 2 consecutive depths → **DIVERGED**, stop, use best depth
- If d = max_depth → stop, use current
**Artifact:** `recursion_log` with RSM and SI at each depth.
**CEI Component:** Recursive Audit (weight 5.0 per level)
---
### 8 — INTERFERENCE: Receive and Provide
**Receive from prior skills:**
| From Skill | Interference | Use |
|------------|--------------|-----|
| CLARIFY | Readiness | If NOT_READY → force MINIMAL budget |
| CONDUCTOR | Task profile | Auto-populate calculus dimensions |
| THRESHOLD | Reversal cost | If Category 1 → force MINIMAL |
**Provide to later skills:**
| To Skill | Interference | Minimum ADS Gain |
|----------|--------------|------------------|
| DEEP-THINK | Shallow score → skip if ≥ 0.67 | +0.15 |
| CONDUCTOR | Depth budget → skill selection | +0.15 |
| THRESHOLD | Reversibility → category | +0.10 |
**Artifact:** `interference_log` (received and provided).
---
## When NOT to Use This Skill
SHALLOW should NOT activate when:
- The user explicitly asks for deep analysis
- The task involves security, safety, or financial consequences
- The decision is hard to reverse (architecture, contract, data model)
- The user has expressed confusion or uncertainty
- This is iteration N>2 on the same problem (depth may finally be warranted)
**The default should still be appropriate depth.** SHALLOW is a targeted counter to depth bias — not a replacement for sound judgment.
---
## The Deeper Purpose
The depth-skills library is optimized for depth. But cognitive investment should match consequence. SHALLOW provides the calibration: a **formal calculus** that routes low-stakes tasks to fast answers. It's not anti-thinking — it's proportional-thinking. **Now it's quantified (shallow score), certified (proportionality ), and verified (gates).** The skill exists because the library's other skills make over-analysis easy. SHALLOW makes under-analysis acceptable — when it's appropriate.
---
## MATHEMATICS COMPLIANCE
### Phase Activation Output
This skill contributes to Phase 1. It outputs:
```json
{
"phase": 1,
"skill": "shallow",
"artifact_completeness": 0.XX,
"gate_pass_rate": 0.XX,
"external_validity": 0.XX,
"human_eval": 0.XX,
"phase_activation": 0.XX
}
```
### CEI Components
- Artifacts produced: depth_budget_calculus (3.0), budget_allocation (2.0), shallow_delivery (1.0), proportionality_ (2.0)
- Total complexity weight: 8.0
### SI Components
- C2/C3 assumptions: 0
- Contrarian viable: false
- Fatal attacks: 0
- Boundary violations: 0
- Assumption reversals: 0
- Proportionality violations: [count if actual ADS deviates from target]
### Cognitive Traces Produced
- [x] Depth Budget Calculus
- [x] Proportionality
- [ ] Decision Landscape
### Gates Verified
- [x] V1 [ ] V2 [ ] V3 [ ] V4 [ ] V5 [x] V6
- [ ] Human Eval [ ] Held-Out [ ] Ensemble
### Required Interferences
- From CLARIFY: Readiness → if NOT_READY force MINIMAL
- From CONDUCTOR: Task profile → auto-populate calculus
- From THRESHOLD: Reversal cost → if Category 1 force MINIMAL