find-cause
Wrap investigation requests with evidence-chain discipline. Use when the user asks to find out why something happens, look into something, research a root cause, debug an issue, or investigate unexpected behavior. Transforms vague investigation requests into reproducible-proof investigations. Invoke with /find-cause <description of what to investigate>.
Evaluation results
82%
41%Transaction Balance Discrepancy
Evidence chain protocol
Capability matrix present
0%
0%
Matrix section labels
0%
0%
CLAIM field used
0%
100%
EVIDENCE field used
0%
100%
VERIFIED field used
0%
100%
DEPENDS ON field used
0%
100%
Script executed
100%
100%
Valid evidence types only
100%
75%
QUESTION section
0%
60%
SUCCESS CRITERIA MET section
0%
80%
EVIDENCE CHAIN section
0%
100%
ROOT CAUSE section
100%
100%
UNVERIFIED ITEMS section
0%
100%
No prohibited words
100%
100%
Without context: $0.3252 · 1m 23s · 16 turns · 21 in / 4,868 out tokens
With context: $0.7738 · 2m 51s · 32 turns · 648 in / 9,955 out tokens
92%
32%Deployment Guard Never Blocks
Reproduce before source analysis
Script executed first
13%
100%
Complete command captured
100%
100%
Complete stdout captured
100%
100%
Exit code recorded
100%
100%
Constraint classification stated
0%
0%
Source line citation
0%
100%
Specific lines quoted
100%
100%
SYMPTOM documented
75%
100%
MECHANISM documented
75%
100%
PROXIMATE CAUSE documented
62%
100%
ROOT CAUSE documented
100%
100%
Without context: $0.1702 · 49s · 9 turns · 14 in / 2,819 out tokens
With context: $0.5597 · 1m 58s · 27 turns · 32 in / 6,646 out tokens
63%
22%Missing Warnings in Log Analysis Report
Interpretation formulation and success criteria
Two or more interpretations
100%
100%
Are you asking X framing
0%
0%
Interpretations are falsifiable
100%
100%
Selected interpretation stated
100%
25%
QUESTION field present
0%
57%
SUCCESS CRITERIA section present
0%
100%
Success criteria includes reproduction item
0%
100%
Success criteria includes root cause formula
0%
100%
Execution mode stated
0%
0%
Constraint classification
0%
0%
Script executed
100%
100%
Without context: $0.2484 · 1m 2s · 13 turns · 18 in / 3,938 out tokens
With context: $0.6472 · 2m 39s · 27 turns · 31 in / 8,886 out tokens
27%
-1%Discount Code Activation Investigation
Unbound constraint safety check batching
Unbound classification
66%
100%
Safety check heading
0%
0%
DESTRUCTIVE OPERATIONS item
0%
0%
SANDBOX item
0%
0%
MISSING INPUTS item
0%
0%
OVERSIGHT LEVEL item
0%
0%
BLIND SPOTS item
0%
0%
Questions batched together
0%
0%
Does not proceed to external systems
100%
50%
Bound analysis completed
100%
100%
Without context: $0.4882 · 2m 31s · 18 turns · 22 in / 8,955 out tokens
With context: $1.2241 · 5m 16s · 43 turns · 46 in / 16,886 out tokens
96%
78%Non-Deterministic Pipeline Output Investigation
Step 0 capability matrix discovery
Diagnostic commands run
16%
100%
Agent directory checked
0%
100%
MCP server check
0%
60%
Capability matrix present
16%
100%
System tools label
0%
100%
MCP servers label
0%
100%
Agents label
0%
100%
Sandbox options label
0%
100%
Matrix precedes investigation
33%
100%
Script actually executed
83%
100%
Without context: $0.8238 · 5m 2s · 25 turns · 26 in / 17,675 out tokens
With context: $1.2344 · 5m 52s · 35 turns · 38 in / 20,544 out tokens
88%
38%Flaky CI Test Investigation
Unverified claim marking and evidence gap handling
VERIFIED: no entries present
50%
100%
Gap description for unverified claims
91%
100%
UNVERIFIED ITEMS section populated
0%
100%
No prohibited hedging words
100%
100%
No gap-filling theories
75%
50%
Local test execution
80%
100%
Source code cited with line numbers
0%
100%
DEPENDS ON: none for reproduction
0%
100%
Suspect claim flagged
37%
62%
No causality without evidence
50%
50%
Without context: $1.3104 · 6m 11s · 40 turns · 43 in / 21,878 out tokens
With context: $1.5192 · 6m 11s · 46 turns · 50 in / 22,280 out tokens
- Repository
- Jamie-BitFlight/claude_skills
- Commit
fd243f9
- Evaluated
- Agent
- Claude Code
- Model
- Claude Sonnet 4.6
Table of Contents
Is this your skill?
If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.