Audit Context Building runs at the start of an audit, threat model or architecture review. It dispatches one subagent per function and collects compact records of what must always be true, what the function takes on faith and which calls it relies on.
The output is a dossier plus one file per function. It names no vulnerabilities and rates no severity. It marks assumptions that nothing in the code enforces and keeps an open questions list for the hunting phase.
What You Get
- A dossier file with an index, rules that span several functions, who can reach what, and where complexity clusters
- One record per function listing invariants with the line that shows them, assumptions with what establishes them, and calls with what each needs
- A list of assumptions marked nothing found, where the code relies on something no code enforces
- An open questions list to carry into the bug hunting phase
Common Use Cases
- Starting an audit of unfamiliar smart contract or C code
- Preparing a threat model before anyone looks for bugs
- Revisiting findings nobody could judge because the system map was missing
- Onboarding onto a service by mapping function contracts first
What Makes It Unique
It deliberately stops short of verdicts. The skill forbids naming vulnerabilities, suggesting fixes or rating severity, and requires every claim to cite a line or become an open question. When two records disagree, both are quoted side by side. The workflow forces subagents to return structured records, so analysis prose never floods the main context.
Under The Hood
The skill is a short instruction file that sends the work to a workflow command and a function analyzer subagent shipped in the same plugin. It lists Workflow, Task, Read, Grep and Glob as allowed tools. The subagents read the target source and write analysis files to disk under an audit-context folder, one dossier and one file per function. It does not run shell commands or reach the network. Format and domain notes are in resource files it links.