The Agent Skills Directory

[DATA_EXPOSURE]: The incident response procedure in modules/incident-response.md recommends capturing the environment state using the command env > /tmp/env_snapshot_$(date +%s).txt. Because environment variables often store sensitive information like API keys, cloud credentials, and database passwords, writing them in plaintext to a shared temporary directory (/tmp) constitutes a local data exposure risk.
[INDIRECT_PROMPT_INJECTION]: The skill acts on untrusted data from project lockfiles (uv.lock, requirements.txt) to identify compromised packages. A maliciously crafted lockfile containing injection payloads in package names or versions could attempt to manipulate the agent's logic or trigger command injection during the scanning process.
Ingestion points: Dependency lockfiles (uv.lock, requirements.txt) and project configuration (pyproject.toml) read by the auditing scripts.
Boundary markers: None present in the skill instructions or code snippets to delimit untrusted data.
Capability inventory: The skill executes shell utilities (grep, find, rg, uv, make) to perform its auditing and remediation tasks.
Sanitization: The provided Python snippets and shell commands do not implement validation or escaping for the data extracted from lockfiles before processing it.
[COMMAND_EXECUTION]: The skill provides and encourages the use of several shell commands (grep, find, ripgrep, uv, make) to audit the local environment and manage dependencies. These commands interact directly with the file system based on paths and package patterns defined in the skill's modules.

supply-chain-advisory