The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes semgrep and npx to perform code analysis and result merging. These operations are restricted to the local environment and a dedicated, user-approved output directory.
[EXTERNAL_DOWNLOADS]: The skill identifies and clones third-party Semgrep rulesets from highly reputable security organizations including Trail of Bits, Decurity, and HashiCorp. These are used to augment standard detection capabilities and are documented as a requirement for a thorough audit.
[DATA_EXFILTRATION]: The instructions mandate the use of --metrics=off for every Semgrep command to prevent the transmission of telemetry or sensitive code metadata to external servers.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted codebases which could theoretically contain malicious patterns intended to trigger false positives or exploit the scanner.
Ingestion points: Reads all code files and project manifests (package.json, pyproject.toml) within the target directory.
Boundary markers: Implements a mandatory 'Hard Gate' at Step 3, requiring explicit user approval of the scan target and rulesets before any execution occurs.
Capability inventory: Includes filesystem read/write (scoped to the output directory), network access for cloning rulesets, and shell command execution for the Semgrep CLI.
Sanitization: Results are processed through structured SARIF/JSON formats, reducing the risk of control-flow manipulation via scan findings.

semgrep