The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The script scripts/golden_diff.py programmatically retrieves the user's GitLab API token by executing the command glab config get token --host gitlab.com. This sensitive credential is used to authenticate requests to the GitLab API.
[DATA_EXFILTRATION]: The scripts/golden_diff.py utility transmits rendered PDF page images (PNG format) to GitLab's upload API (https://gitlab.com/api/v4/projects/{project_id}/uploads) and posts the resulting image references to Merge Request comments. While documented as a feature, this facilitates the movement of potentially sensitive document content to an external service.
[COMMAND_EXECUTION]: Multiple scripts within the skill utilize subprocess.run to execute external binaries on the host system:
scripts/golden_diff.py executes git (diff, rev-parse, show), gs (GhostScript), montage (ImageMagick), diff-pdf, and curl.
scripts/verify_field_alignment.py executes gs (GhostScript) to render PDF pages.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect injection as it processes untrusted PDF data.
Ingestion points: PDF files are read and parsed in scripts/add_row.py, scripts/inspect_fields.py, scripts/verify_field_alignment.py, and scripts/apply_content_stream_replacements.py using the pikepdf and pypdf libraries.
Boundary markers: None identified; the skill operates directly on content streams and annotation metadata.
Capability inventory: The skill has capabilities to write files (pdf.save, writer.write), execute shell commands (various tools via subprocess.run), and perform network operations (curl via subprocess and urllib.request.urlopen in scripts/golden_diff.py).
Sanitization: The skill lacks explicit sanitization or filtering of external content before processing it for structural changes or rendering.

ac-editing-acroforms