The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill retrieves a QUEST_TOKEN by reading a sensitive environment file at a hardcoded absolute path: /home/ubuntu/projects/fellowship-dev/claude-buddy/.env. While the path appears associated with the vendor's project structure, hardcoding absolute paths to sensitive configuration files is a security risk.
[DYNAMIC_EXECUTION]: The skill uses python3 -c to execute dynamically generated code. In Step 4, the shell variables $REPORT_TITLE and $GROUP are interpolated directly into the Python source string. This lacks proper escaping or parameterization, allowing for potential Python code injection if these variables contain single quotes or other escape sequences.
[COMMAND_EXECUTION]: The skill performs multiple system operations including file writing, directory resolution via git rev-parse, and network transmission via curl. The curl command specifically sends data to a local API (127.0.0.1:4242) using the token harvested from the .env file.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect injection as it processes untrusted inputs (such as report titles and job manifests) without boundary markers or sanitization.
Ingestion points: Inputs provided via the --jobs flag and environment variables like $REPORT_TITLE and $GROUP used in Step 4.
Boundary markers: Absent; the content is processed directly.
Capability inventory: Subprocess calls to git, python3, and curl; file write operations.
Sanitization: Absent; variables are directly interpolated into script strings.

write-report