The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection (Category 8) by processing untrusted data from multiple project sources.
Ingestion points: Reads repository configuration and context files such as package.json, pyproject.toml, requirements.txt, .github/workflows/*.yml, and CLAUDE.md (found in SKILL.md).
Boundary markers: Absent. The instructions do not specify the use of delimiters or provide "ignore embedded instructions" warnings for the content read from the repository.
Capability inventory: The skill has the capability to execute shell commands for tool verification and generate user-facing reports containing installation instructions.
Sanitization: Absent. Content extracted from repository files is interpolated into the analysis and final report without explicit validation or escaping.
[COMMAND_EXECUTION]: The skill performs dynamic command execution to verify the availability of detected development tools.
Evidence: Executes commands including bunx vitest --version, bunx biome --version, tsc --version, pytest --version, ruff --version, and cargo clippy --version (detailed in SKILL.md Step 2c).
Context: These commands are limited to version verification for auditing purposes.
[EXTERNAL_DOWNLOADS]: The skill provides installation recommendations for external software.
Evidence: Recommends the command curl -fsSL https://bun.sh/install | bash for the Bun runtime (found in references/tool-matrix.md).
Context: This recommendation targets a well-known service (Bun) and is presented as guidance in the output report rather than being executed automatically by the skill.

agent-dx