threshold — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited threshold (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Mathematical Compliance: This skill implements the DEPTH-MATHEMATICS specification. It computes and outputs ADS phase activation, CEI components, SI components, cognitive traces (reversal cost function, commitment , early warning signals), and verifies Gates V1-V6. It provides the final commitment verdict for the Depth .
Not all decisions deserve the same depth. A CSS color choice and a database primary key choice are not the same decision — but without this skill, the model treats them with roughly equal cognitive investment.
This skill creates the gate between thinking and committing. It measures the reversal cost, enforces proportional exploration, and requires an exit plan before any high-cost commitment. Now it quantifies reversal cost as a function, produces verifiable commitment s, and defines early warning signals.
You are about to give a quick, confident answer to a decision that:
Uniform depth — treating all decisions with the same level of analysis — systematically under-invests in the decisions that matter most.
New failure mode in v2.0: Reversal cost handwaving — "hard to undo" without quantification. Commitment theater — gates that look strict but lack enforcement. Early warning absence — no observable signals to detect a wrong decision before it's too late.
Read thinking_parameters:
require_reversal_cost_function (default true) — quantify reversal cost mathematicallyrequire_commitment_ (default true) — produce verifiable commitmentrequire_early_warning (default true) — define observable early warning signalsrequire_gate_enforcement (default true) — enforce all gates before commitmentNEW IN v2.0 — Reversal cost is not a category. It's a function.
THRESHOLD CLASSIFICATION
────────────────────────────────────────
Decision: [what is being decided — one sentence]
Reversal Cost Function R(t):
R(t) = C_fixed + C_variable(t) + C_opportunity(t) + C_risk(t)
Where:
C_fixed = [one-time costs: migration tooling, data transformation, downtime]
C_variable(t) = [ongoing costs: parallel run, dual-write, performance penalty]
C_opportunity(t) = [foregone benefits: what you can't do while reversing]
C_risk(t) = [probability × impact of reversal failure]
Parameters:
t = time since commitment [days]
R(0) = [immediate reversal cost]
R(30) = [cost at 30 days]
R(90) = [cost at 90 days]
R(365) = [cost at 1 year]
R(∞) = [permanent cost — if never fully reversible]
Reversal Timeline: [minutes / hours / days / weeks / months / never]
Blast Radius: [self / team / users / ecosystem]
Category: [1 / 2 / 3 / 4] — derived from R(∞)
Category 1 — TRIVIAL (R(∞) < $1K equivalent, minutes-hours):
Variable names, file organization, CSS values, config
Category 2 — MODERATE (R(∞) $1K-$100K, hours-days):
Indexes, internal API shapes, library versions, service boundaries
Category 3 — EXPENSIVE (R(∞) $100K-$1M, days-weeks):
Database schemas, public API contracts, auth systems, data formats
Category 4 — CATASTROPHIC (R(∞) > $1M or impossible, weeks-never):
Primary keys, data deletion, published standards, architecture at scale
When uncertain between two categories, pick the higher one.
The cost of overthinking a reversible decision is wasted time.
The cost of underthinking an irreversible one is permanent damage.
────────────────────────────────────────Artifact: The reversal cost function with parameters and category.
CEI Component: Reversal Cost Function (weight 4.0)
Based on the classification:
DEPTH REQUIREMENT
────────────────────────────────────────
Category 1 — STANDARD:
→ 1 approach is sufficient. Move fast.
→ Skip remaining steps. Deliver.
ADS Target: 0.20
Category 2 — ELEVATED:
→ 2 approaches minimum before committing.
→ Document rollback path (1-2 sentences).
→ Run G1-G3 gates (Step 3).
ADS Target: 0.45
Category 3 — MAXIMUM:
→ 3 approaches minimum (use DIVERGE skill).
→ Full ADVERSARY review of chosen approach.
→ Written rollback plan (Step 4).
→ Run ALL gates G1-G6 (Step 3).
ADS Target: 0.70
Category 4 — MAXIMUM-PLUS:
→ All Layer 0 skills activated (DEEP-THINK → DIVERGE → ADVERSARY).
→ No commitment without explicit user confirmation.
→ Written rollback plan with timeline and dependencies (Step 4).
→ Run ALL gates. No exceptions.
ADS Target: 0.85
────────────────────────────────────────Artifact: The depth requirement with ADS targets.
Before any answer ships at Category 2+, these gates must pass. Write the result for each required gate:
TERMINATION GATES
────────────────────────────────────────
G1 — Problem fidelity:
Can I restate the user's intent (not just their words)?
[PASS — restatement: ___] / [FAIL — unclear because ___]
G2 — Alternative coverage:
Were at least [2/3] materially different approaches considered?
[PASS — approaches: ___] / [FAIL — only [N] considered]
CIM Check: [CIM score ≥ 0.5? PASS/FAIL]
G3 — Assumption exposure:
Are critical assumptions named and rated (C2/C3 from EXCAVATE)?
[PASS — assumptions listed] / [FAIL — assumptions not surfaced]
G4 — Failure-mode analysis: (Category 3+ only)
Are failure paths identified, not just the happy path?
[PASS — failures: ___] / [FAIL — only happy path]
NEGATIVE-SPACE Check: [failure taxonomy complete? PASS/FAIL]
G5 — Reversibility plan: (Category 3+ only)
Is the exit plan written (Step 4)?
[PASS — plan written] / [FAIL — no exit plan]
Reversal Cost Function: [R(t) defined? PASS/FAIL]
G6 — Decision rationale: (Category 3+ only)
Can I explain why the chosen option beats alternatives under
THESE SPECIFIC constraints?
[PASS — because: ___] / [FAIL — generic reasoning]
PROVENANCE Check: [EC ≥ 0.5? PASS/FAIL]
G7 — Early warning signals: (Category 3+ only, NEW)
Are observable early warning signals defined?
[PASS — signals: ___] / [FAIL — no signals]
Monitoring Plan: [dashboards/alerts defined? PASS/FAIL]
G8 — Commitment : (Category 4 only, NEW)
Is the commitment written (Step 5)?
[PASS — written] / [FAIL — no ]
User Confirmation: [explicit confirmation received? PASS/FAIL]
────────────────────────────────────────Any FAIL → return to the relevant step and complete it before shipping. The gate exists because coherence is not completion and confidence is not evidence.
Artifact: The gate results. Failed gates block delivery.
CEI Component: Termination Gates (weight 3.0)
EXIT PLAN
────────────────────────────────────────
If this decision turns out wrong, reversal requires:
Reversal Cost Function: R(t) = [from Step 1]
Steps:
1. [specific action — e.g., "disable feature flag X"]
2. [specific action — e.g., "run migration script Y in reverse"]
3. [specific action — e.g., "restore database from snapshot Z"]
4. [specific action — e.g., "notify stakeholders via channel W"]
5. [specific action — e.g., "verify system health via dashboard V"]
Timeline: [estimated reversal time at t=0, t=30, t=90]
R(0): [time]
R(30): [time]
R(90): [time]
Data Risk: [what could be lost or corrupted during reversal]
Data Loss Risk: [NONE / LOW / MEDIUM / HIGH]
Corruption Risk: [NONE / LOW / MEDIUM / HIGH]
Recovery Procedure: [how to recover if corruption occurs]
Dependencies: [what/who must coordinate]
Internal Teams: [list]
External Vendors: [list]
User Communication: [required?]
Early Signals: [what observable sign indicates this decision is going wrong]
Signal 1: [metric/threshold] — Detection: [dashboard/alert] — Action: [what to do]
Signal 2: [metric/threshold] — Detection: [dashboard/alert] — Action: [what to do]
Signal 3: [metric/threshold] — Detection: [dashboard/alert] — Action: [what to do]
Monitoring Plan:
Dashboards: [list]
Alerts: [list]
Review Cadence: [daily/weekly/monthly]
Owner: [role/person]
────────────────────────────────────────If you cannot write the exit plan — you don't understand the decision well enough to make it. Return to exploration.
Artifact: The exit plan with reversal cost function, timeline, data risk, dependencies, early signals, and monitoring plan.
CEI Component: Exit Plan (weight 4.0)
NEW IN v2.0 — Verifiable, auditable commitment record.
COMMITMENT
────────────────────────────────────────
ID: cc_<timestamp>_<hash>
Decision: [what is being committed to]
Category: [4 — CATASTROPHIC]
Reversal Cost: R(∞) = [value] — [Category 4 justification]
Chosen Option: [name]
Alternatives Considered: [list with CIM scores]
Option A: [name] — CIM: 0.XX — Rejected because: [specific]
Option B: [name] — CIM: 0.XX — Rejected because: [specific]
Gate Results:
G1: PASS — [restatement]
G2: PASS — [approaches, CIM: 0.XX]
G3: PASS — [assumptions]
G4: PASS — [failures]
G5: PASS — [exit plan, R(t)]
G6: PASS — [rationale, EC: 0.XX]
G7: PASS — [early signals]
G8: PASS — [this ]
Early Warning Signals:
1. [metric] ≥ [threshold] → [action]
2. [metric] ≥ [threshold] → [action]
3. [metric] ≥ [threshold] → [action]
Monitoring:
Dashboard: [URL]
Alerts: [list]
Review Cadence: [weekly]
Owner: [role/person]
User Confirmation:
Confirmed By: [user identifier]
Timestamp: [ISO 8601]
Acknowledgment: "I understand this decision is Category 4 (CATASTROPHIC reversal cost).
I have reviewed the exit plan, early warning signals, and alternatives.
I authorize this commitment."
Expiration: [date or condition for re-evaluation]
────────────────────────────────────────Artifact: commitment_ — the final gate for Category 4 decisions.
CEI Component: Commitment (weight 5.0)
require_early_warning = true)NEW IN v2.0 — Observable signals that a decision is going wrong, defined BEFORE commitment.
EARLY WARNING SIGNALS
────────────────────────────────────────
For the chosen option, define signals that would indicate the decision is failing:
Signal 1: [Name]
Metric: [specific measurable metric — e.g., "p99 latency > 500ms"]
Threshold: [specific value — e.g., "> 500ms for 5 consecutive minutes"]
Detection: [how measured — e.g., "Datadog dashboard 'API Latency'"]
Lead Time: [how early this signal appears before catastrophic failure — e.g., "2 weeks"]
False Positive Rate: [estimated — e.g., "5%"]
Action: [specific response — e.g., "trigger rollback procedure Phase 1"]
Owner: [who responds]
Signal 2: [Name]
...
Signal 3: [Name]
...
Signal Validation:
Each signal must be:
- Observable (measurable without human judgment)
- Specific (threshold is a number, not "feels slow")
- Actionable (triggers a defined response)
- Early (appears before irreversible damage)
- Low noise (false positive rate < 20%)
Composite Early Warning Index:
EWI = Σ (Signal_Weight × Signal_Active) / Σ Signal_Weight
If EWI > 0.7 → ESCALATE (trigger review/rollback)
────────────────────────────────────────Artifact: early_warning_signals with metrics, thresholds, detection, lead times, actions, and composite index.
CEI Component: Early Warning Signals (weight 3.0)
#### 7.1 Phase Activation (for ADS)
{
"phase": 5,
"skill": "threshold",
"artifact_completeness": 0.XX,
"gate_pass_rate": 0.XX,
"external_validity": 0.XX,
"human_eval": 0.XX,
"phase_activation": 0.XX
}#### 7.2 CEI Components
{
"cei_components": {
"threshold_classification": 2.0,
"depth_requirement": 1.0,
"termination_gates": 3.0,
"exit_plan": 4.0,
"reversal_cost_function": 4.0,
"commitment_": 5.0,
"early_warning_signals": 3.0,
"total_complexity_weight": 22.0,
"output_tokens": N,
"wall_time_seconds": T,
"cei": 0.XX
}
}#### 7.3 SI Components
{
"si_components": {
"c2_c3_count": 0,
"contrarian_viable": false,
"fatal_attacks": 0,
"boundary_violations": 0,
"assumption_reversals": 0,
"si_total": 0
}
}#### 7.4 Cognitive Traces
Reversal Cost Function:
{
"type": "reversal_cost_function",
"decision": "...",
"category": 3,
"function": "R(t) = C_fixed + C_variable(t) + C_opportunity(t) + C_risk(t)",
"parameters": {
"C_fixed": 10000,
"C_variable_daily": 500,
"C_opportunity_daily": 2000,
"C_risk_probability": 0.1,
"C_risk_impact": 50000
},
"values": {
"R(0)": 10000,
"R(30)": 85000,
"R(90)": 265000,
"R(365)": 1000000,
"R(inf)": 1500000
},
"blast_radius": "users",
"category": 3
}Commitment : (as defined in Step 5)
Early Warning Signals:
{
"type": "early_warning_signals",
"signals": [
{"name": "Latency Spike", "metric": "p99_latency", "threshold": 500, "unit": "ms",
"detection": "Datadog", "lead_time_days": 14, "fpr": 0.05, "action": "rollback_phase1"},
{"name": "Error Rate", "metric": "error_rate", "threshold": 0.01, "unit": "ratio",
"detection": "PagerDuty", "lead_time_days": 7, "fpr": 0.02, "action": "investigate"}
],
"composite_ewi": 0.0,
"escalation_threshold": 0.7
}#### 7.5 Depth Contribution
{
"skill": "threshold",
"version": "2.0",
"metrics": {"ads_contribution": 0.XX, "cei": 0.XX, "si": 0, "ec": 0.XX},
"gates_verified": {"V1": true, "V2": true, "V3": true, "V4": true, "V5": true, "V6": true},
"limitations": ["..."]
}MANDATORY — Verify these gates PASS (all categories as applicable):
| Gate | Check | Pass Condition |
|---|---|---|
| V1 | Problem Fidelity | Restatement captures user intent |
| V2 | Alternative Coverage | ≥2/3 approaches with CIM ≥ 0.5 |
| V3 | Assumption Exposure | C2/C3 assumptions named and rated |
| V4 | Failure-Mode Analysis | Failure paths identified (NEGATIVE-SPACE) |
| V5 | Reversibility Plan | Exit plan + R(t) written |
| V6 | Decision Rationale | Specific rationale + EC ≥ 0.5 |
| V7 | Early Warning Signals | Observable signals with thresholds defined |
| V8 | Commitment | Category 4: + user confirmation |
If any gate FAILS: Return to relevant step and fix. Do not deliver.
Receive from prior skills:
| From Skill | Interference | Use |
|---|---|---|
| TEMPORAL | Regret scenarios | Inform reversal cost function |
| NEGATIVE-SPACE | Critical silences | Define early warning signals |
| PROVENANCE | EC score | Gate G6 threshold |
| DIVERGE | CIM score | Gate G2 threshold |
| ADVERSARY | Fatal attacks | Inform failure-mode analysis |
Provide to later skills:
| To Skill | Interference | Minimum ADS Gain |
|---|---|---|
| CONDUCTOR | Final commitment verdict | +0.15 |
| META-LEARNING | Decision patterns for policy learning | +0.10 |
Artifact: interference_log (received and provided).
This skill creates automatic depth-scaling. Trivial decisions get fast answers. Irreversible decisions get deep exploration. The gateway prevents the most expensive class of error: confident, fast, wrong, and permanent. Now it quantifies reversal cost as a function, produces verifiable commitment s, and defines observable early warning signals. Stopping is a decision — this skill ensures it's a considered, measurable, and auditable one.
This skill contributes to Phase 5. It outputs:
{
"phase": 5,
"skill": "threshold",
"artifact_completeness": 0.XX,
"gate_pass_rate": 0.XX,
"external_validity": 0.XX,
"human_eval": 0.XX,
"phase_activation": 0.XX
}~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.