The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill facilitates the download of the 'opentaint' analysis tool and its associated installation scripts from the vendor's official domain (opentaint.org) and the '@seqra' NPM organization.
[REMOTE_CODE_EXECUTION]: Installation instructions provided to the agent involve piping remote shell scripts (install.sh and install.ps1) directly into system interpreters (bash and iex). While these resources originate from the vendor's infrastructure, this execution pattern is highly sensitive.
[COMMAND_EXECUTION]: The skill performs various system operations, including building project models, executing the analyzer, and managing Docker containers using docker compose to verify vulnerabilities via Proof-of-Concept (PoC) scripts. It also includes automated cleanup logic that terminates processes and containers registered in .opentaint/tracking/poc-servers.yaml.
[PROMPT_INJECTION]: The skill has an indirect prompt injection surface due to its data processing pipeline.
Ingestion points: Untrusted data enters the agent context via SARIF report files and dependency metadata extracted from the analyzed project.
Boundary markers: Analysis findings are stored in YAML tracking files and passed to triage subagents without explicit boundary markers or 'ignore embedded instructions' warnings.
Capability inventory: The skill possesses capabilities for file system writes, subprocess execution, and network service management (via Docker).
Sanitization: The sarif-to-findings.py script extracts message text from SARIF results and interpolates them into YAML files without specific escaping or validation to prevent instruction injection.

appsec-agent