The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes several local shell commands to manage session history, including invoking Node.js scripts for listing and preprocessing transcripts, and running embedded Python scripts for extraction and assembly of restored documents. It also uses git log to retrieve project history based on session timestamps.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface (Category 8) as it reads and re-injects untrusted data from historical session transcripts into the current conversation context. Ingestion points: Transcripts are read from ~/.claude/projects/ and preprocessed cache files are read into the context via Step 4 and Step 5e. Boundary markers: The skill uses headers such as [Context restored by /continue] and [Session:{sid} {ISO} L{n}] as delimiters, but it does not provide explicit instructions for the agent to ignore potentially malicious commands embedded in the restored text. Capability inventory: The skill has capabilities for file system access (read/write), shell command execution (Node.js, Python, Git), and environment variable access. Sanitization: While the project path is sanitized into a hash via sed, the content of the transcripts themselves is not sanitized or filtered for instructions before being loaded into the context.

continue