Contradiction Detector

Detects and eliminates contradictions between agent instruction files that cause hallucinations and silent misbehavior. Use when: (1) any agent behaves incon...

Security Scan

The skill's requested binaries, file access, and runtime instructions match its stated purpose of scanning agent instruction files for contradictions; nothing requested is disproportionate or unexplained.

Purpose & Capability

Name/description (detect contradictions across AGENTS.md, SOUL.md, HEARTBEAT.md, MEMORY.md, cron payloads, skills, hooks) aligns with required binaries (openclaw, python3, find, cat) and declared workspace/openclaw.json read access. The requested capabilities are expected and proportional. Note: the SKILL.md assumes a specific filesystem layout (examples use /opt/ocana/openclaw/workspace and /opt/ocana/openclaw/openclaw.json); confirm these paths match your installation.

Instruction Scope

SKILL.md explicitly describes discovery, cross-referencing, reporting, and requiring user confirmation before applying fixes. It instructs the agent to read many internal instruction sources (files, cron prompts, hook scripts, skills). This broad read-only scope is necessary for the stated audit purpose, but it will access potentially sensitive internal files and inline cron prompts.

Install Mechanism

Instruction-only skill with no install spec or downloaded code — lowest-risk install model. No archives, third-party package installs, or external download URLs are present.

Credentials

No environment variables or credentials requested. Declared configPaths/readAccess (openclaw.json, workspace, workspace/skills, workspace/hooks) are appropriate for the audit task. No unrelated secrets or external API keys are required.

Persistence & Privilege

always:false (no forced inclusion). disable-model-invocation:false is normal (skill can be invoked autonomously), but SKILL.md explicitly requires user confirmation before applying fixes and forbids autonomous destructive changes (removing files, altering cron schedules, changing ownership). The skill does not request system-wide config writes in its instructions.

Assessment

This skill appears coherent and performs a local read-only audit of agent instruction sources. Before installing, consider: (1) It needs the openclaw CLI and filesystem access to your workspace/openclaw.json — confirm those paths match your environment. (2) It will read many internal files (AGENTS.md, HEARTBEAT.md, MEMORY.md, cron inline prompts, hook scripts, skills); those files can contain sensitive data — ensure you are comfortable granting the skill read access. (3) The skill states it will present fixes as diffs and will not apply them without explicit confirmation; nevertheless, always back up files before accepting automated edits. (4) If you rely on a non-standard layout or different openclaw.json location, expect the discovery commands to require adaptation. If any of the above is unacceptable, do not enable the skill or run it in a controlled/read-only environment first.

Contradictions between agent instruction files are the #1 cause of silent misbehavior in multi-agent OpenClaw setups. Files are loaded non-deterministically across sessions — the only fix is making all files agree.

Prerequisites

• openclaw CLI — listing agents and cron jobs
• python3 — parsing JSON config
• find, cat — standard POSIX utilities
• Filesystem access to the OpenClaw workspace directory

Audit Procedure

Step 1 — Discovery

Enumerate every instruction source: all agents, their workspace files (AGENTS.md, SOUL.md, HEARTBEAT.md, MEMORY.md, IDENTITY.md, USER.md), all cron jobs with inline prompts, all installed skills, and any hook scripts.

→ Detailed procedure: references/discovery.md

Build a complete inventory per agent before proceeding. Missing even one source means contradictions slip through.

Step 2 — Cross-reference for contradictions

Systematically check every pair of instruction sources per agent against 12 known contradiction patterns:

#	Pattern	Risk	Most common trigger
1	Format duplication	HIGH	Same output format defined in 2+ files independently
2	Cron prompt override	HIGH	Cron contains stale instructions; HEARTBEAT.md fix is invisible
3	Routing mismatch	HIGH	Different session/channel targets for same output
4	Persona/tone conflict	MED	SOUL.md vs AGENTS.md communication style
5	Trigger/schedule mismatch	MED	Cron frequency vs HEARTBEAT.md timing
6	Memory routing conflict	MED	Same data type sent to different memory locations
7	Dead/orphaned rules	MED	Old rule not removed after newer rule added elsewhere
8	Cross-agent assumption mismatch	MED	Sender expects X, receiver does Y
9	Implicit vs explicit behavior	LOW-MED	Gap filled differently by different files
10	Skill-agent conflict	LOW-MED	Skill instructions override agent rules when triggered
11	Stale MEMORY.md directives	LOW	"From now on" in MEMORY.md contradicts current AGENTS.md
12	Environment/config drift	LOW	Hardcoded paths, model names, or URLs that changed

→ Full detection procedures and examples: references/contradiction-patterns.md

Step 3 — Report and fix

Produce a structured finding for each contradiction (agent, severity, pattern, evidence with exact quotes, impact, proposed fix). Present all fixes as diffs and wait for user confirmation before applying.

→ Report format, severity guidelines, and fix protocol: references/reporting-and-fixes.md

Post-fix verification

After every fix: re-read both files, verify agreement, and confirm no new contradictions were introduced.

Scheduling (optional)

This skill works as a manual on-demand invocation. Optionally add to HEARTBEAT.md or a cron job for periodic runs.