The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Node.js scripts (list-sessions.js, preprocess.js, and restore.js) located within the plugin's root directory to perform session listing, caching, and context slicing.
[DYNAMIC_EXECUTION]: The skill uses shell heredocs to dynamically generate and execute three separate Python scripts. These scripts are used to extract user turn headers from cache files, identify relevant turns for topic-based restoration, and batch extract full message content from session JSONL files.
[INDIRECT_PROMPT_INJECTION]: The skill restores arbitrary text from previous session transcripts into the active conversation, which could contain malicious instructions from past interactions.
Ingestion points: Reads session transcripts from ~/.claude/projects/ and ~/.codex/sessions/, as well as compact.txt files from ~/.claude/super-token-saver-data/.
Boundary markers: Relies on specific header strings like [Session:...] and User: "..." to distinguish between turns, but does not use delimiters to prevent the LLM from following instructions within the restored text.
Capability inventory: The skill allows the agent to execute shell commands, local JavaScript scripts, dynamically generated Python code, and read/write to the local filesystem.
Sanitization: There is no evidence of sanitization, filtering, or escaping applied to the content retrieved from session transcripts before it is loaded into the prompt context.

s-continue