The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes a local PowerShell script scripts/find-artifacts.ps1 and uses the git status command to identify files that have been staged or modified recently.
[DATA_EXPOSURE]: The skill reads file metadata (names, paths, sizes, and timestamps) from the local filesystem. This is limited by a whitelist of user-facing extensions (e.g., .md, .pdf, .xlsx, .png) and a blacklist of sensitive or internal directories (e.g., .git, .idea, node_modules, pycache). It does not read file contents or expose secrets.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted filenames and metadata which are subsequently rendered as Markdown links.
Ingestion points: File metadata returned by find-artifacts.ps1.
Boundary markers: Absent.
Capability inventory: PowerShell and Git command execution; no file-write or network capabilities.
Sanitization: No explicit sanitization of filenames is performed before rendering. The risk is considered low as the agent only performs formatting and the skill lacks the capabilities required for exfiltration or persistence.

artifact-radar