The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses local Python scripts to orchestrate the discovery workflow. Scripts like run_semgrep_scout.py and run_discovery_quality_gate.py utilize subprocess.run to invoke the Semgrep CLI (for local scanning) or to execute other internal utility scripts for data processing and validation. These executions are limited to local files and installed tools.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because its primary purpose is to ingest and analyze untrusted source code from external repositories.
Ingestion points: Repository source files are read during profiling (profile_repositories.py) and retrieval (build_context_packs.py).
Boundary markers: Code hits are wrapped in Markdown code blocks within "context packs" with structured headers.
Capability inventory: The agent has access to file system reading and local command execution via its toolset.
Sanitization: The methodology relies on structured AI triage, adversarial R2 review, and quality gates to identify and filter malicious or misleading content within the analyzed repositories.
[DYNAMIC_EXECUTION]: Static analysis detected the use of pickle.loads and dynamic execution patterns. These are used strictly within the skill's own Semgrep rules and Python detection logic to identify vulnerabilities in the code being audited, or within unit tests (tests/test_workflow_optimizations.py) to verify the scanner's efficacy. The skill does not execute these primitives on data in its own runtime environment.

repo-bug-finder