Knowledge conflict resolution

Detect conflict between knowledge sources used in the same task and resolve it via source precedence, with escalation when precedence is not enough.

Canonical skill profile. This page is generated from the repository contract. The linked SKILL.md remains the source of authority.

Jump to

• Overview
• When to use
• When not to use
• Core moves
• Expected output
• Verification
• Handoff signals
• Pairs well with
• Anti-patterns

View the canonical contract on GitHub.

Overview

Use this skill when two or more sources selected by the knowledge resolver disagree on a decision-relevant point. It applies the source-precedence-policy, records the resolution in the audit log, and escalates when precedence cannot settle the conflict.

When to Use

• A task’s resolved context pack contains sources that disagree.
• A skill output would change depending on which source is treated as authoritative.
• Two mandatory sources (e.g. privacy vs. accessibility) appear to conflict.

When NOT to Use

• Selecting sources in the first place (handled by the resolver).
• Evaluating whether a single source can be registered (use knowledge-source-evaluation).
• Checking exposure risk of restricted excerpts (use restricted-context-check).

Core Moves

1. State the conflict precisely: which sources, which decision point, which positions.
2. Locate each source in the precedence tiers.
3. Apply precedence; record the prevailing source and the suppressed sources.
4. Re-derive the affected decision under the prevailing source.
5. If precedence does not settle it (same tier, no tie-break, mandatory-vs-mandatory, deliverable-breaking), escalate to human review.

Optional Modules

• Tier tie-break — apply, in order: authority level → scope specificity → recency → supersedes.
• Mandatory clash protocol — generate a structured review request when two mandatory sources collide.
• Suppressed-source carry-forward — preserve the lower source as context in the output, not as authority.

Activation Triggers

• Resolver flags conflicts_detected in the context pack.
• A reviewer questions which source governed a decision.
• A mandatory source is present alongside an interpretive or evidential source on the same point.

Expected Output

conflict_resolution:
  conflict_id: <uuid>
  task_id: <task-id>
  between:
    - <pack_id@version>
    - <pack_id@version>
  topic: <decision point in one line>
  positions:
    - source: <pack_id@version>
      position: <one line>
    - source: <pack_id@version>
      position: <one line>
  resolved_by: source_precedence_policy | escalation
  prevailing_source: <pack_id@version | null>
  suppressed_sources: [<pack_id@version>, ...]
  human_review_required: <bool>
  human_review_reason: <if true>
  effect_on_deliverable: <short>

Verification

• Conflict is described precisely enough that a reviewer can replay the reasoning.
• Precedence chosen matches the tier mapping in source-precedence-policy.
• Suppressed sources are preserved as context, not erased.
• Escalation fired whenever precedence did not settle the conflict.
• Escalation matches the mandatory triggers in human-review-criteria (mandatory-source conflict, regulatory risk, high-impact decision).
• The conflict record is carried forward per logs-and-handoffs-policy without exposing restricted excerpts.

Handoff Signals

• Pass the structured conflict record to the next agent, preserving restrictions per logs-and-handoffs-policy.
• If escalated, do not produce a final deliverable until review returns.

Pairs Well With

• knowledge-source-evaluation
• restricted-context-check
• feature-value-governance (and any knowledge-aware skill)

Anti-patterns

• “Splitting the difference” between a mandatory and an interpretive source.
• Dropping the lower-precedence source silently instead of preserving it as context.
• Treating recency as a primary tie-break (it is a tertiary one).
• Resolving in the skill output without writing an audit entry.