The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/office/soffice.py dynamically generates C source code, writes it to a temporary file, and compiles it into a shared object library using gcc. This library is then injected into the LibreOffice process using the LD_PRELOAD environment variable to intercept and modify system socket calls. This technique is used to bypass environment restrictions but represents a high-risk behavior.
[COMMAND_EXECUTION]: The skill makes extensive use of shell command execution through subprocess.run across multiple scripts:
scripts/office/soffice.py executes the compiler gcc.
scripts/thumbnail.py executes soffice and pdftoppm for file conversion.
scripts/office/validators/redlining.py executes git diff for document comparison.
[METADATA_POISONING]: The LICENSE.txt file included with the skill claims copyright by 'Anthropic, PBC', which directly contradicts the metadata listing the author as 'likas07'. This is deceptive metadata that could lead users to misjudge the provenance and safety of the skill.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a significant attack surface for indirect prompt injection as it processes untrusted PowerPoint files and passes their content into prompts for subagents:
Ingestion points: scripts/office/unpack.py reads user-provided .pptx archives and their internal XML components.
Boundary markers: Explicit boundary markers or instructions to ignore embedded commands are absent in the subagent prompt template provided in SKILL.md.
Capability inventory: The skill has powerful capabilities including arbitrary command execution and dynamic code compilation.
Sanitization: There is no evidence of sanitization for the text or XML content extracted from source presentations before it is used in subsequent agent operations.

pptx