evidence-chain
Skill1 starsChanged 5 months agoApache-2.0View on GitHub
Multi-source evidence chain analysis. Extracts evidence from multiple independent sources, cross-validates, detects conflicts, evaluates confidence, and outputs analysis reports with root cause judgments. Use when: (1) analyzing client complaints against system records to find truth, (2) cross-validating data from 2+ sources (complaints, alerts, SLA agreements, operation logs), (3) detecting contradictions between sources, (4) producing confidence-scored root cause judgments. Triggers: evidence chain, cross-validation, multi-source analysis, conflict detection, confidence assessment, root cause inference, fault diagnosis analysis, complaint verification, alert correlation, evidence chain analysis.
In this repository4 other files it gives its agents.
- data-aggregatorSkill
- nl2-querySkill
- scoring-engineSkill
- visualization-rendererSkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 2 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 38
If any component fails to load, execute degradation strategy (see "Error Handling")
- Distinctive: line 241
All archive records must include: analyst, analysis time, evidence source list
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Evidence-Chain
Core philosophy: Not single-source RAG answering from a knowledge base, but multi-source evidence chain — ordinary RAG answers from a knowledge base; advanced Skills cross-validate from multiple sources, detect conflicts, and evaluate confidence.
Component Orchestration
This Skill chains 4 components that must execute in strict sequence; each stage's output is the next stage's input:
Info-Extractor ──→ Data-Analyst ──→ Report-Generator ──→ Archive-Manager
Phase 1 Phase 2 Phase 3 Phase 4
Evidence Cross-validation Report Archiving
extraction + inference generation & audit trailExecution Rules:
- Each component must be loaded via the
skilltool before execution; skipping or merging is prohibited - Components pass intermediate results as structured JSON objects (see output contracts per Phase below)
- If any component fails to load, execute degradation strategy (see "Error Handling")
Detailed orchestration protocol: references/orchestration.md.
Input Parameters
Required
| Parameter | Type | Description |
evidence_sources | JSON array | Each element contains source_name, source_type, content |
analysis_question | string | Core question, e.g. "Is there a persistent service quality issue with the network service?" |
source_type enum: complaint_report / system_alert / contract_term / operation_log / other
Optional
| Parameter | Type | Default | Description |
conflict_sensitivity | enum | medium | high / medium / low |
confidence_threshold | 0-1 | 0.7 | Below this value, root cause is annotated "⚠️ Needs further verification" |
output_format | enum | markdown | markdown / json / table |
Phase 1 — Info-Extractor: Evidence Extraction
Invocation: Load info-extractor skill via skill tool, execute extraction for each source
Input: Each element in evidence_sources
Processing:
- Iterate
evidence_sources, call Info-Extractor for each source - Extract structured fields: time, event, values, objects, location, etc.
- Annotate each piece of evidence with source type and initial confidence
| Source Type | Initial Confidence | Attribute |
| complaint_report | ⭐⭐ (0.4) | Subjective statement |
| system_alert | ⭐⭐⭐⭐ (0.8) | Objective data |
| contract_term | ⭐⭐⭐ (0.6) | Authoritative reference |
| operation_log | ⭐⭐⭐⭐ (0.8) | Objective data |
| other | ⭐⭐⭐ (0.6) | Pending assessment |
Phase 1 Output Contract (passed to Phase 2):
{
"extracted_evidences": [
{
"evidence_id": "E-01",
"source_name": "Source Name",
"source_type": "complaint_report",
"extracted_fields": { "time": "...", "event": "...", "value": "..." },
"initial_confidence": 0.4,
"raw_summary": "One-line summary"
}
]
}Phase 2 — Data-Analyst: Cross-Validation and Inference
Invocation: Load data-analyst skill via skill tool, use Phase 1 output as analysis input, execute cross-validation and inference per the sub-steps below
Input: Phase 1 output_phase1 object
Sub-steps:
2a - Timeline Alignment
- Sort
extracted_evidenceschronologically - Identify multi-source records of the same event, annotate time deltas
- Output:
timeline_eventsarray
2b - Conflict Detection
- Compare descriptions of the same event from different sources
- Adjust detection granularity per
conflict_sensitivity - Each conflict annotated: conflict ID, type, description, involved sources, difference analysis
- Conflict detection rules: references/conflict-rules.md
2c - Confidence Assessment
- Formula:
composite_confidence = source_confidence × cross_validation_score × time_decay- Cross-validation score: proportion of independent sources confirming the same fact (0-1)
- Time decay: evidence older than 30 days decays by 0.95^months
2d - Root Cause Inference
- Three dimensions: Earliest Occurrence / Downstream Dependency / High Frequency
- Output: suspected root cause + confidence level (High/Medium/Low) + numeric value + basis
- Below
confidence_threshold→ annotate "⚠️ Needs further verification"
Phase 2 Output Contract (passed to Phase 3):
{
"timeline_events": [...],
"conflicts": [
{
"conflict_id": "C-01",
"type": "value_conflict",
"description": "...",
"sources_involved": ["E-01", "E-03"],
"difference": "delta=1 occurrence",
"possible_reason": "..."
}
],
"confidence_scores": [
{ "evidence_id": "E-01", "source_confidence": 0.4, "cross_validation": 0.7, "time_decay": 1.0, "final_confidence": 0.28 }
],
"root_cause": {
"suspected": "...",
"confidence_level": "High",
"confidence_value": 0.85,
"basis": ["Earliest occurrence", "Downstream dependency", "High frequency"],
"needs_further_verification": false
},
"recommended_actions": [
{ "priority": "🔴 Urgent", "action": "...", "linked_evidence": ["E-04"] },
{ "priority": "🟡 Concurrent", "action": "...", "linked_evidence": ["E-01","E-05"] }
]
}Phase 3 — Report-Generator: Generate Report
Invocation: Load report-generator skill via skill tool, generate report from Phase 2 analysis results per template
Input: Phase 2 output_phase2 object + Phase 1 output_phase1 object
Output Format: Selected per output_format parameter
Module 1: Evidence List Table
| # | Source | Source Type | Time | Key Content | Confidence |
Module 2: Conflict Detection Section
Each conflict: conflict ID + type + description + involved sources + difference analysis
Module 3: Root Cause Judgment
Suspected root cause + confidence (High/Medium/Low) + numeric value + basis + ⚠️ marker (if applicable)
Module 4: Recommended Actions
| Priority | Recommended Action | Linked Evidence |
| 🔴 Urgent | ... | Evidence #n |
| 🟡 Concurrent | ... | Evidence #n |
| 🟢 Deferred | ... | Evidence #n |
Module 5: Metadata
Analysis time / Evidence count / Conflict count / Max confidence
Phase 3 Output Contract (passed to Phase 4):
{
"report_content": "Full report text (markdown/json/table)",
"report_metadata": {
"analysis_time": "2026-06-20T...",
"evidence_count": 7,
"conflict_count": 2,
"max_confidence": 0.85
}
}Phase 4 — Archive-Manager: Archiving
Invocation: Load archive-manager skill via skill tool for archiving
Input: Phase 3 output_phase3 object
Archive Contents:
- Analysis process: Phase 1 extraction results + Phase 2 analysis results
- Analysis conclusion: Phase 3 full report
- Archive metadata: Analyst, analysis time, evidence source list
Pre-Archive Anonymization Required:
- Client name → "Client A"
- Contract IDs retained but financial details hidden
Component Load Failure Degradation Strategy
| Component | Fallback when load fails |
| Info-Extractor | Manually extract structured fields per source type, annotate initial confidence |
| Data-Analyst | Manually execute Phase 2 sub-steps (timeline alignment / conflict detection / confidence assessment / root cause inference) per SKILL.md rules |
| Report-Generator | Manually generate report per Phase 3's 5-module template in SKILL.md |
| Archive-Manager | Save archive contents as local .md file, filename includes timestamp and analysis question |
Degraded reports must annotate in metadata: "degraded": true, "degraded_components": ["info-extractor"]
Error Handling
| Condition | Behavior |
| Source count = 0 | Error: "At least 2 evidence sources are required for cross-validation" |
| Source count = 1 | Degrade to single-source analysis + warning: "Only 1 source available; cross-validation not possible; conclusion confidence is limited" |
| All evidence missing time info | Skip Phase 2a + report annotation "Missing time dimension; chronological analysis not possible" |
| sensitivity=high but no conflicts | Output "No conflicts detected; all source descriptions are consistent" |
| Root cause confidence < threshold | Annotate "⚠️ Needs further verification" + list evidence that needs supplementation |
Security Requirements
- Client complaint content must not be archived to public knowledge bases before anonymization
- SLA contract terms are confidential; must be anonymized during archiving
- Root cause judgments must include confidence level; low-confidence conclusions must not be used as sole basis for actions
- All archive records must include: analyst, analysis time, evidence source list
Demo Data
Full 3-source evidence chain example: references/demo-data.md.
Self-Check Checklist
- Can correctly extract structured fields from 3 different source types
- Can detect conflicts between client statements and system records
- Can produce confidence-scored root cause judgments
- Can distinguish 🔴 Urgent / 🟡 Concurrent / 🟢 Deferred action priorities
- Anonymized data correctly hides client names
- Archive records include complete evidence source list
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

