multi-stance-audit-fanout
Skill220 starsChanged 20 days agoMITView on GitHub
Use after automated EVM review has left a small Solidity target needing human-style logic coverage—prepare one source bundle for eight isolated audit stances, then, as the appointed coordinator, deduplicate their returned leads and detect composite chains. Specialists package and consolidate; Chrono alone dispatches the stance packets.
In this repository179 other files it gives its agents.
- AGENTS.mdAGENTS.md
- chrono/CLAUDE.mdCLAUDE.md
- CLAUDE.mdCLAUDE.md
- accessible-media-authoringSkill
- agentic-safety-auditSkill
- agent-prompt-engineeringSkill
- audio-event-map-authoringSkill
- auto-scaffoldSkill
- chain-constructSkill
- chain-construct-smart-contractSkill
- claim-verificationSkill
- code-reachability-auditSkill
- code-review-loopSkill
- color-theorySkill
- conversation-designSkill
- copy-refinementSkill
- cosmos-sdk-audit-checklistSkill
- cross-chain-bridge-auditSkill
- cross-file-relationship-synthesisSkill
- data-flow-traceSkill
- dedup-prior-art-checkSkill
- defi-invariant-checkSkill
- dependency-cycle-auditSkill
- dependency-health-triageSkill
- detection-as-codeSkill
- diff-aware-semgrep-scanSkill
- differential-reviewSkill
- dimensional-analysis-checkSkill
- dual-level-retrievalSkill
- durable-nonce-exploitationSkill
- erc1271-revert-data-checkSkill
- error-based-sstiSkill
- evm-audit-flowSkill
- figma-implement-designSkill
- findings-filterSkill
- forensic-timeline-authoringSkill
- game-design-fundamentalsSkill
- game-mechanics-balancingSkill
- head-tailSkill
- incident-response-runbookSkill
- interactive-audio-designSkill
- interface-ambiguity-checkSkill
- keyword-clusteringSkill
- knowledge-base-integrationSkill
- known-advisory-backport-checkSkill
- layered-analysis-loopSkill
- level-design-patternsSkill
- locale-adaptationSkill
- narrative-structureSkill
- osint-platform-auditSkill
- parser-differential-route-confusionSkill
- persuasive-technical-writingSkill
- platform-complianceSkill
- player-engagement-psychologySkill
- pre-audit-threat-modelSkill
- program-rubric-lookupSkill
- read-only-reentrancy-checkSkill
- requirements-elicitationSkill
- review-severity-ladderSkill
- rule6-rights-gateSkill
- The first 60 shown.
The file
We split this skill into the rules it gives an agent
7 rules. About 0 are common, 5 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 17
NEVER self-spawn, launch a model CLI, or dispatch sub-agents — that is Chrono's job.
- Very specific: line 76
Composite chain pass: If finding A's output-state precondition feeds B AND combined impact is strictly worse than either alone, emit Chain: …
- Distinctive: line 88
Do NOT self-spawn the stances — surface the fan-out to Chrono.
- Distinctive: line 89
Do NOT inline source into stance prompts repeatedly — bundle once, reference path.
- Distinctive: line 90
Do NOT collapse stances into a single mega-prompt.
- Very specific: line 91
Do NOT promote LEADs on stance-count alone — [stances: 2+] does not override a concrete refutation by another stance with code evidence.
- Distinctive: line 93
Do NOT reason about deployer intent — evaluate what the code *allows*.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
<!-- inspired by pashov/skills:solidity-auditor (MIT); recast Chrono-orchestrated -->multi-stance-audit-fanout
Stance-based audit fan-out. Eight specialist personas read the same Solidity source through
different mental models; findings are then deduplicated by group key with composite-chain
detection. Distinct from evm-audit-flow (tool pipeline) — this fans out human-style review
across audit lenses to catch logic bugs no detector finds.
Orchestration model (read first). This is a Chrono-orchestrated fan-out. Workers
NEVER self-spawn, launch a model CLI, or dispatch sub-agents — that is Chrono's job. A
security-analyst worker does NOT run the eight stances itself. Instead this skill defines (a)
the stance roster + briefs, (b) the source bundle, and (c) the dedup/promotion method, so a
worker can either surface the fan-out as a need (## NEEDS FROM CHRONO: dispatch the 8
stance packets over <bundle>) or, when acting AS the Chrono-appointed coordinator, dedup the
returned stance outputs. Chrono dispatches N stance packets on the board rail; a coordinator
role dedups and writes the single consolidated result.
When to use
- A contract has passed
evm-audit-flowstatic + symbolic passes but still needs logic-bug coverage no detector catches. - The target is small enough (≤2k nSLOC) for eight readers to cover end-to-end.
- You want stance diversity (8 lenses, 1 model) cheaply, before paying for model diversity (a full multi-model fan-out).
- Skip when a fresh equivalent report already exists; do not duplicate.
Stance roster
Each stance gets the SAME source bundle plus its own role brief. The eight stances are:
- vector-scan — sweep known vuln classes (reentrancy, oracle manip, signature replay, storage collision, delegatecall hijack, slippage, sandwich, MEV).
- math-precision — rounding direction, decimal scaling, fixed-point overflow, division before multiplication, dust accumulation, precision loss across token decimals.
- access-control — role surface, two-step transfer correctness, initializer reentrancy, modifier ordering, bypass via internal-call routing.
- economic-security — incentive design, MEV extraction, fee accrual paths, redemption griefing, donation attacks, rebalance arbitrage, liquidation race conditions.
- execution-trace — pick the top 3 fund-flow paths, single-step from entry to settlement, note every state read/write/external call along the way; chase reachability not patterns.
- invariant — derive invariants the protocol should hold (
totalSupply == sum(balances),kfor AMMs, collateralization ratio, share-to-asset monotonicity), then look for paths that break each one. - periphery — ERC-20/721/777 quirks, fee-on-transfer, rebasing, missing return values,
safeTransfergaps,permitfront-running, callback re-entry surfaces. - first-principles — ignore checklists; ask "what is this contract trying to do, and what would I do as the attacker if I had infinite time?" Capture novel/unnamed bug classes.
Order of operations
- Bundle source (worker or coordinator). Build a single
source.mdwith every in-scope.solbehind a### pathheader and a fenced block. Skipinterfaces/,lib/,mocks/,test/,*.t.sol,*Test*.sol,*Mock*.sol. Record line count. - Fan out (Chrono). Chrono dispatches one board packet per stance — stance name, the
stance brief from the roster, the bundled source, and the output schema below. Eight
independent workers; no worker spawns another. If you are a worker and the fan-out has not
been dispatched, surface it under
## NEEDS FROM CHRONO— do not attempt to launch it. - Output schema per stance. Each stance returns FINDING and LEAD entries:
FINDING|LEAD: <Contract>.<function> — <bug-class> group_key: <Contract>|<function>|<bug-class> confidence: 0-100 rationale: 1-3 sentences trace: <numbered call/state steps if applicable> - Deduplicate by group_key (coordinator). Exact-match merge first. Then merge synonymous
bug-class tags that share the same contract/function (e.g.
slippage≈price-impact). Annotate each merged finding with[stances: N]. - Composite chain pass. If finding A's output-state precondition feeds B AND combined
impact is strictly worse than either alone, emit
Chain: [A] + [B]atconfidence = min(A, B). Most reviews surface 0–2 chains. - Promote LEAD → FINDING. Promote at confidence 75 if either: (a) a complete exploit
chain is traced in source, OR (b)
[stances: 2+]agree even when one stance demoted. Never promote on stance-count alone without source-level confirmation. - Hand off. Send the deduplicated finding set to the multi-model FP-filter and then
impact-validator(G1–G4) before recording — stance diversity catches different bugs than model diversity does, and reachability alone does not pay.
Anti-patterns
- Do NOT self-spawn the stances — surface the fan-out to Chrono; workers never launch sub-agents.
- Do NOT inline source into stance prompts repeatedly — bundle once, reference path.
- Do NOT collapse stances into a single mega-prompt; the diversity comes from role isolation.
- Do NOT promote LEADs on stance-count alone —
[stances: 2+]does not override a concrete refutation by another stance with code evidence. - Do NOT reason about deployer intent — evaluate what the code allows.
Recording (chrono-vault)
The task packet's injected memory contract owns the exact call shape, sequence, and fields - see
wirework-reflect. Do not copy a record(...) example or add fields (including source_task) from
memory; the server binds them, and a baked example violates the run's authenticated schema. Memory is
best-effort telemetry and never gates the work. What is worth recording here is the task-specific
outcome: stance count, raw/merged finding counts, chains, leads promoted, bundle path.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

