The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a bundled shell script, scripts/detect-stack.sh, to identify the project's package manager and installed toolchain. This script uses standard utilities such as grep, find, and dirname to perform local project introspection.
[DATA_EXFILTRATION]: No network operations or data exfiltration vectors were identified. The skill's output is restricted to a local file within a .plans directory in the project's root.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a data ingestion surface that could be targeted via project configuration files.
Ingestion points: The agent reads content from package.json, tsconfig.json, and existing .plans/setup-ts-project-config/plan.md files during the reconciliation process.
Boundary markers: Absent. The instructions do not specify the use of delimiters or 'ignore' instructions when reading these external files.
Capability inventory: The skill can execute a local shell script (scripts/detect-stack.sh) and has file-write access to the specific plan path.
Sanitization: Absent. There are no instructions for sanitizing or escaping the data read from project files before it is included in the plan output.
Impact: The risk is mitigated by the skill's design, which generates a text-based plan for human review rather than executing the commands it proposes.
[EXTERNAL_DOWNLOADS]: The skill provides templates and dependency lists for well-known development tools (e.g., ESLint, Prettier, Vitest) but does not perform any automated downloads or installations itself.

setup-ts-project-config