/zeroize-audit

name: zeroize-audit
description: "Detects missing zeroization of sensitive data in source code and identifies zeroization removed by compiler optimizations, with assembly-level analysis, and control-flow verification. Use for auditing C/C++/Rust code handling secrets, keys, passwords, or other sensitive data."
allowed-tools: Read Grep Glob Bash Write Task AskUserQuestion mcp__serena__activate_project mcp__serena__find_symbol mcp__serena__find_referencing_symbols mcp__serena__get_symbols_overview

zeroize-audit

When to Use

• Auditing cryptographic implementations (keys, seeds, nonces, secrets)
• Reviewing authentication systems (passwords, tokens, session data)
• Analyzing code that handles PII or sensitive credentials
• Verifying secure cleanup in security-critical codebases
• Investigating memory safety of sensitive data handling

When NOT to Use

• General code review without security focus
• Performance optimization (unless related to secure wiping)
• Refactoring tasks not related to sensitive data
• Code without identifiable secrets or sensitive values

---

Execution Contract

Use the agent environment's available tools for file access, shell commands, worker execution, and user input. The `allowed-tools` entry and role-file frontmatter are legacy host metadata; they do not define another environment's tool names or grant permissions there.

• **Resolve resources.** Set `baseDir` to the absolute directory containing this `SKILL.md`. Confirm the linked [input schema](schemas/input.json) and [orchestrator prompt](prompts/task.md) exist there. `{baseDir}`, `{workdir}`, and `{{input}}` are notation for values the orchestrator must substitute, not environment variables or automatic host expansion. Resolve relative Markdown links against the directory of the file containing them; resolve `{baseDir}/…` paths against the skill directory. Pass resolved paths to workers and shell commands.
• **Launch a role.** Read the role prompt linked in the Agent Architecture table. When the environment has a registered worker for the role, use its `zeroize-audit:<role-name>` identifier (for example, `zeroize-audit:0-preflight`); otherwise create an independent worker with the Markdown body of that prompt, the current phase's instructions, and its parameter values. Always include the resolved `baseDir` and, once created, `workdir`, even when a phase table does not repeat them. Resolve resource references using the rules above before delegation. Worker file/search/shell actions use the environment's actual tool schemas and permissions. The role's name identifies its instructions, not a required dispatch API or model.
• **Wait for results.** Retain each worker's handle and wait through the environment's completion mechanism. A launch acknowledgement is not completion. Read the phase's required artifacts before updating its completion state or starting a dependent phase. On worker failure or unavailable dispatch, apply only that phase's documented fallback; otherwise record the failure in the existing run state and report the affected phase as blocked.
• **Concurrency.** Respect available concurrency and phase barriers. Run workers sequentially when parallel execution is unavailable. Per-TU compiler workers write to separate directories and may run concurrently.
• **Collect decisions.** Use the environment's user-input facility or ask in the conversation. When a phase requires an answer, wait for it; do not substitute a default or elapsed time for the user's decision. Preserve the validation and verification outcomes when recording any accepted or rejected PoC result.

How to Run

On a request like "audit this crate for secrets left in memory" or "check that this C library actually wipes its keys":

1. **Collect inputs.** Map the request onto the Inputs table below (full schema: [schemas/input.json](schemas/input.json)). `path` is required, plus at least one of `compile_db` (C/C++) or `cargo_manifest` (Rust); if neither is given or derivable from the repo, ask the user, because preflight stops the run without one. Leave all other fields at their defaults unless the user says otherwise. 2. **Read the [orchestrator prompt](prompts/task.md)**, substituting the collected inputs for its `{{placeholder}}` values. You act as the orchestrator it describes: it defines state recovery, the phase loop, early termination, and error handling. Read the [shared instructions](prompts/system.md) alongside it for the shared working-directory layout and the agent error protocol every phase depends on. 3. **Execute its phase loop.** Run Phases 0-7 sequentially. Before each phase, read that phase's workflow file from `{baseDir}/workflows/phase-{N}-{name}.md` and follow its Preconditions, Instructions, State Update, and Error Handling sections. Each workflow specifies the role prompt and parameters to use under the Execution Contract above. Honor the per-phase skip conditions and the early-termination rules in task.md. 4. **Return the report** (Phase 8, inline): read `{workdir}/report/final-report.md` and return its contents as the skill output.

To resume an interrupted run: if a `workdir` is known from prior context, read `{workdir}/orchestrator-state.json` and continue from its `current_phase` instead of starting at Phase 0 (see the Recovery section of task.md).

---

Purpose

Detect missing zeroization of sensitive data in source code and identify zeroization that is removed or weakened by compiler optimizations (e.g., dead-store elimination), with mandatory LLVM IR/asm evidence. Capabilities include:

• Assembly-level analysis for register spills and stack retention
• Data-flow tracking for secret copies
• Heap allocator security warnings
• Semantic IR analysis for loop unrolling and SSA form
• Control-flow graph analysis for path coverage verification
• Runtime validation test generation

Scope

• Read-only against the target codebase (does not modify audited code; writes analysis artifacts to a temp