The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill clones third-party Semgrep rulesets from various GitHub repositories, including those maintained by security consulting firms and researchers. This is a core feature for expanding security coverage beyond default rules.
The skill requires a "HARD GATE" user approval (Step 3) before any rulesets are cloned or scans are executed.
URLs are validated via regex in the shell script to ensure they are legitimate HTTPS git repositories.
[COMMAND_EXECUTION]: The skill executes the Semgrep CLI and associated shell scripts (run-scans.sh) to perform analysis.
A dedicated function has_unsafe_char is used to sanitize target and output paths, rejecting any inputs containing shell-active characters like backticks, dollar signs, or quotes.
Telemetry is explicitly disabled using the --metrics=off flag on every semgrep invocation to prevent data leakage.
[DYNAMIC_EXECUTION]: The skill uses uv run to execute a local Python script (merge_sarif.py) for consolidating scan results.
The script is executed without project dependencies (--no-project) and performs file-based merging without shelling out to external tools.
[INDIRECT_PROMPT_INJECTION]: As a static analysis tool, the skill ingests and processes untrusted codebases. This represents an indirect prompt injection surface.
Ingestion points: The codebase target directory is read by the Semgrep engine.
Boundary markers: The scan results are written to a dedicated output directory and are not directly interpolated into the agent's system instructions.
Capability inventory: The skill has capabilities for network access (via git clone), file system modification (within the output directory), and subprocess execution.
Sanitization: Shell injection protection is implemented for all user-supplied paths, and downloaded rule files are pruned to remove non-rule YAML content that might cause scanner instability.

semgrep