agentleFS
yuzhaopeng-upyuzhaopeng-up/teleagent-skills/skills/evidence-chain/SKILL.md

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.

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
  1. Distinctive:

    If any component fails to load, execute degradation strategy (see "Error Handling")

    line 38
  2. Distinctive:

    All archive records must include: analyst, analysis time, evidence source list

    line 241

≈ “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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

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 trail

Execution Rules:

  • Each component must be loaded via the skill tool 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

ParameterTypeDescription
evidence_sourcesJSON arrayEach element contains source_name, source_type, content
analysis_questionstringCore 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

ParameterTypeDefaultDescription
conflict_sensitivityenummediumhigh / medium / low
confidence_threshold0-10.7Below this value, root cause is annotated "⚠️ Needs further verification"
output_formatenummarkdownmarkdown / 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:

  1. Iterate evidence_sources, call Info-Extractor for each source
  2. Extract structured fields: time, event, values, objects, location, etc.
  3. Annotate each piece of evidence with source type and initial confidence
Source TypeInitial ConfidenceAttribute
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_evidences chronologically
  • Identify multi-source records of the same event, annotate time deltas
  • Output: timeline_events array

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

#SourceSource TypeTimeKey ContentConfidence

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

PriorityRecommended ActionLinked 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

ComponentFallback when load fails
Info-ExtractorManually extract structured fields per source type, annotate initial confidence
Data-AnalystManually execute Phase 2 sub-steps (timeline alignment / conflict detection / confidence assessment / root cause inference) per SKILL.md rules
Report-GeneratorManually generate report per Phase 3's 5-module template in SKILL.md
Archive-ManagerSave 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

ConditionBehavior
Source count = 0Error: "At least 2 evidence sources are required for cross-validation"
Source count = 1Degrade to single-source analysis + warning: "Only 1 source available; cross-validation not possible; conclusion confidence is limited"
All evidence missing time infoSkip Phase 2a + report annotation "Missing time dimension; chronological analysis not possible"
sensitivity=high but no conflictsOutput "No conflicts detected; all source descriptions are consistent"
Root cause confidence < thresholdAnnotate "⚠️ 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

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.