You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
feat: Devil's Advocate (Challenger) agent charter template + auto-trigger #598
Agents overstate findings and produce unchallenged analysis. In our squad, agents claimed 75-90% token savings when the real number was 20-55%. Without a challenger, wrong recommendations get implemented.
Proposal
Add an optional Challenger agent template to the framework:
Charter template:.squad/templates/agents/challenger.md — role is to verify claims, run counter-hypotheses, check math, flag confidence levels (Verified/Unverified/Contradicted)
Spawn integration: Coordinator can auto-spawn challenger before any architecture decision or when a claim exceeds a threshold (e.g., 'saves 75%')
Iterative retrieval pattern: Max 3 investigation cycles, must cite evidence for every verdict
Output format: Per-claim verdict table with evidence links
Prior Art
Field-tested as 'Q' agent across 200+ issues. Caught: inflated metrics, fabricated config references, wrong bottleneck assumptions. False positive rate ~15% (challenged things that were actually correct).
Deliverables
Agent charter template
Ceremony integration (auto-trigger before architecture decisions)
changed the title [-]feat: Devil's Advocate (Challenger) agent template[/-][+]feat: Devil's Advocate (Challenger) agent charter template + auto-trigger[/+]on Mar 24, 2026
Revision after checking main: The Reviewer Rejection Protocol already exists in squad.agent.md — that covers post-work review. What's missing is a pre-decision challenger pattern: auto-spawning a fact-checker before architecture decisions or when agents make claims with specific metrics. The charter template + ceremony auto-trigger are the gap. Narrowing scope to those.
Resolved by #603 (merged). Challenger agent template + fact-checking skill are now available in templates/agents/challenger.md. Teams can adopt this pattern immediately.
Activity
[-]feat: Devil's Advocate (Challenger) agent template[/-][+]feat: Devil's Advocate (Challenger) agent charter template + auto-trigger[/+]tamirdresher commentedon Mar 24, 2026
Revision after checking main: The Reviewer Rejection Protocol already exists in squad.agent.md — that covers post-work review. What's missing is a pre-decision challenger pattern: auto-spawning a fact-checker before architecture decisions or when agents make claims with specific metrics. The charter template + ceremony auto-trigger are the gap. Narrowing scope to those.
feat(skills): add fact-checking skill\n\nAdds challenger/fact-checkin…
feat(agents): add challenger agent charter template\n\nGeneric Devil'…
feat: add Challenger (Devil's Advocate) agent template
feat(agents): add Challenger / Devil's Advocate agent template + fact…
bradygaster commentedon Mar 26, 2026
Resolved by #603 (merged). Challenger agent template + fact-checking skill are now available in templates/agents/challenger.md. Teams can adopt this pattern immediately.
fix: REPL UX fixes — 10 issues resolved with 52 new tests (github#608)
feat(agents): add Challenger / Devil's Advocate agent template + fact…
feat(agents): add Challenger / Devil's Advocate agent template + fact…
feat: APM integration — squad skill publish/install + apm.yml in init (…
feat(agents): add Challenger / Devil's Advocate agent template + fact…
feat: APM integration — squad skill publish/install + apm.yml in init (…