The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses external command-line utilities to extract text from legacy document formats and Outlook archives found in user backups.
Evidence: SKILL.md contains implementation instructions for using antiword, catdoc, and readpst to parse file content. Notably, readpst is directed to write to /tmp/pst-output, which is outside the skill's declared writes_to directories.
[DYNAMIC_EXECUTION]: The skill employs Python one-liners to perform runtime parsing of archive metadata and modern document structures.
Evidence: SKILL.md specifies the use of python3 -c for validating .pst file headers and extracting text from .docx files by parsing internal XML structures within the ZIP container.
[INDIRECT_PROMPT_INJECTION]: As a tool designed to ingest large amounts of unstructured personal data, the skill is susceptible to instructions embedded within the scanned files.
Ingestion points: Reads content from .mbox, .doc, .docx, and .pst files across multiple user-defined source types (local, Dropbox, B2).
Boundary markers: The skill tracks state in a STATUS.md manifest, but does not define explicit delimiters or instructions to ignore embedded prompts when presenting raw file content to the agent for triage.
Capability inventory: The agent can write to originals/, personal/, and ideas/ directories and execute shell commands to parse files.
Sanitization: The skill logic includes stripping HTML tags for display but lacks a comprehensive sanitization strategy for other potentially malicious text formats.

archive-crawler