The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill performs dynamic code generation and compilation at runtime.
In pptx/scripts/office/soffice.py, the _ensure_shim function writes an embedded C source string to a file in the temporary directory and compiles it into a shared library using gcc via subprocess.run.
The script uses the LD_PRELOAD environment variable to inject this library into the soffice (LibreOffice) process to modify its system call behavior for networking.
[EXTERNAL_DOWNLOADS]: The skill fetches external software binaries during execution.
SKILL.md contains instructions for the agent to run npx playwright install chromium, which downloads browser binaries from a remote repository managed by Microsoft.
[COMMAND_EXECUTION]: The skill executes multiple system-level utilities and shell scripts.
pptx/scripts/office/soffice.py executes gcc for compilation.
pptx/scripts/thumbnail.py and pptx/scripts/office/soffice.py invoke soffice and pdftoppm.
pptx/scripts/office/validators/redlining.py uses git diff for document comparison.
SKILL.md runs a shell script status.sh to check project state.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection through the processing of untrusted project data.
Ingestion points: SKILL.md reads project-specific HTML files (*-r*.html) to perform conversions and visual QA.
Boundary markers: The instructions do not define delimiters for the HTML content or provide warnings to ignore instructions embedded within the processed deck files.
Capability inventory: The skill has broad capabilities including file system access, network-adjacent browser execution via Playwright, and shell command execution.
Sanitization: No evidence of sanitization or filtering of the HTML content is present before it is rendered by Playwright or inspected by the agent.

codeck-export