The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes several local shell scripts (wait-for-review.sh, log-post.sh, show-posts.sh) to manage the review gate workflow. These scripts interact with the file system to read/write drafts and maintain a log file at ~/.claude/posts-log.md.
[INDIRECT_PROMPT_INJECTION]: The workflow involves reading content from ~/Downloads/ immediately before posting it to external platforms. If another process or a malicious actor modifies these files while the agent is waiting, the agent might post unauthorized content. The skill instructions explicitly tell the agent to 'Re-read it from disk immediately before posting' and that 'The file is the source of truth, not the version drafted in the conversation'.
Ingestion points: wait-for-review.sh and subsequent cat or read operations on ~/Downloads/*.md files.
Boundary markers: None; the skill treats the file content as the final text to be posted.
Capability inventory: File read/write via bash scripts, shell execution of bash scripts, background process execution.
Sanitization: None; the scripts use shasum for change detection but do not validate the safety or integrity of the text content itself.

post-review-gate