The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and semantically analyze entire codebases from user-provided paths. This activity exposes the agent to untrusted data that could contain indirect prompt injections designed to manipulate the audit findings or the automated code fixes in 'Phase 2'.
Ingestion points: Source files, configuration, and documentation within the user-provided directory are read during Pass 0 through Pass 3, as specified in SKILL.md.
Boundary markers: There are no explicit boundary markers or instructions provided to the agent to treat ingested file content as purely data, increasing the likelihood of the agent following embedded instructions.
Capability inventory: The skill has the ability to execute shell scripts, run external linters, write to the filesystem (AUDIT.md), and modify existing code files.
Sanitization: No sanitization, escaping, or validation of the ingested code strings is performed before processing.
[COMMAND_EXECUTION]: The skill executes bundled shell scripts (scripts/scan_dead_code.sh, scripts/scan_dependencies.sh, scripts/scan_patterns.sh) to facilitate codebase auditing. It also invokes standard development tools in the execution environment, such as eslint, tsc, pylint, flake8, go vet, and cargo clippy. These executions are intended for static analysis of the target codebase.

iblai-vibe-deslop