trace-sources

name	trace-sources
description	Trace and verify the sources behind claims, code references, and recommendations. Use when: the user invokes /trace-sources, wants to audit where information came from, needs to verify code references exist, or asks 'how do you know that?'. Targets: hallucination, black-box debugging, retrieval failures.
user-invocable	true

Trace Sources — Retrieval Auditability

You are performing a source audit on your most recent response (or the content the user specifies). Every claim needs a traceable origin.

Step 1: Claim Extraction

List every factual claim, code reference, and recommendation in the response. Include:

• Direct assertions ("function X does Y")
• Implicit claims ("this is the standard approach")
• Code references (file paths, function names, variable names, API endpoints)
• Behavioral claims ("this will cause / prevent / fix...")

Step 2: Source Classification

For each claim, assign a source tag:

• [FILE:path:line] — Verified by reading a specific file at a specific location in this session
• [TOOL:name] — Verified by tool output (grep, glob, bash, etc.) in this session
• [USER] — Stated by the user in this conversation
• [INFERRED] — Logically derived from verified sources (state the derivation)
• [TRAINING] — From general knowledge / model training data (not verified in this session)
• [UNVERIFIED] — Cannot trace to any source

Step 3: Verification Pass

For all [FILE] and [TOOL] tagged claims:

• Re-read the referenced files to confirm the claim is accurate
• Check that referenced functions, classes, and variables actually exist
• Verify line numbers are approximately correct
• Confirm that behavioral descriptions match actual code logic

For [TRAINING] tagged claims:

• Flag claims where general knowledge may be outdated
• Flag claims where project-specific behavior might differ from general patterns
• Consider whether verification via codebase search is feasible

Step 4: Unverified Items

For each [UNVERIFIED] claim:

• Can it be verified now? (search the codebase, read a file, etc.)
• If verifiable: do it and update the tag
• If not verifiable: flag it prominently and state what would be needed to verify

Step 5: Audit Summary

Source Audit Results:
  Verified (FILE/TOOL):  X claims
  User-provided:         X claims
  Inferred:              X claims
  Training knowledge:    X claims
  Unverified:            X claims

Verification rate: X%
Items requiring attention: [list]

Highlight any claims that turned out to be wrong during re-verification.