aiskillstore/marketplace / skills/aakashh242/gauntlet
gauntlet
VerifiedEvidence-driven adversarial review and root-cause repair of code, designs, APIs, infrastructure, tests, and agent workflows. Use for deep reviews, red-team reviews, edge-case audits, production-readiness checks, or requests to find and fix defects iteratively. Supports sub-agents or a tracked solo workflow. Not for ordinary implementation or cosmetic edits unless explicitly invoked.
references/root-cause.md
View on GitHub# Root-cause repair, not symptom suppression
## A finding must explain a causal chain
Write the observed symptom, minimum trigger, failing mechanism, enabling condition, and the invariant that should have prevented the failure. Separate observed facts from hypotheses. For a missing requirement, the root may be an undefined contract; do not invent product policy and call its implementation a fix.
Ask a counterfactual: **if this mechanism were corrected, would the failure still be possible through another caller or schedule?** Search for sibling uses of the same primitive and duplicate checks. Bound that search by the actual causal mechanism, not an unrelated full-codebase rewrite.
## Choose the smallest sufficient boundary
Prefer enforcing an invariant at its owner. A database uniqueness rule may need a constraint and a caller-compatible conflict path; a local precheck alone may still race. A retry duplication issue may require an idempotency key plus durable deduplication at the commit boundary, not a sleep. A tenant leak may need tenant-scoped authorization in a shared data access path, not hiding one UI button. The correct design depends on the actual system; these are mechanisms to investigate, not mandatory prescriptions.
Document why the chosen boundary blocks the failure class and what assumptions remain. A root cause can be a single incorrect operator. Do not inflate “root cause” into architectural replacement when a narrow, causally complete correction is safer.
## Test the proposed fix before accepting it
A regression should fail on the original behavior for the intended reason and pass on the repaired behavior. Distinguish a real reproduction from a model of the behavior. When a pre-fix execution would be unsafe or unavailable, label the substitute evidence and the verification gap rather than inventing a failure log.
Test a neighboring instance of the defect class: another caller, another tenant, an alternate encoding, repeated delivery, a different cancellation point, or an independent schedule. Include unaffected valid behavior. Generated tests must derive expectations from the requirement, not simply echo the patched implementation.
## Reject these shortcuts
Suppressing an exception without restoring the invariant; widening a timeout without understanding an unbounded wait; retries without idempotency; deleting or weakening failing tests; hardcoding a single input; treating string prefixes as authorization; compensating for inconsistent writes only at rendering time; swallowing partial failures; silently changing a public contract; upgrading a dependency blindly; introducing a new abstraction unrelated to the defect.
A mitigation can be appropriate during an incident. Label it **mitigation**, record the residual mechanism and expiry/owner, and do not mark the root-cause repair verified. The user must authorize operational incident actions independently.
## Patch acceptance record
Record original trigger and evidence, root cause, affected sites, invariant after the change, exact patch snapshot, defect regression, neighboring-class coverage, compatibility/rollout risks, and verification result. Review the patch itself for a new failure mechanism. When two attempts do not resolve the same defect, revisit the causal model and oracle before trying another modification.
Stop for authorization when the repair requires changing product semantics, widening permissions, a destructive migration, substantial public API breakage, production access, or a broad refactor outside scope. Preserve the evidence and propose the smallest next decision.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |