The Agent Skills Directory

[COMMAND_EXECUTION]: The script article_status.py uses subprocess.run to execute git commands (rev-parse, status, log) and a local helper script last-article.sh. While the helper script is expected in a sibling directory (../write-review/scripts/), executing shell commands on user-provided or inferred article paths presents a standard command execution surface.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources that may contain malicious instructions.
Ingestion points: Reads frontmatter from index.md files and scans Claude Code project transcripts (~/.claude/projects/*/*.jsonl).
Boundary markers: No explicit delimiters or instructions are provided to the agent to ignore potential commands embedded in the article content or transcript logs.
Capability inventory: The skill can execute local shell commands (via git) and trigger the execution of other skills like write-review.
Sanitization: The script uses regex patterns to extract specific fields but does not sanitize the resulting strings before presenting them to the agent.
[DATA_EXPOSURE]: The skill reads Claude Code transcripts to track the history of writing skill usage. These files contain the user's full interaction history with the AI, which is sensitive information, although it is processed locally in this context.

write-status