The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple shell commands including node, sed, awk, git, and python3 to identify projects, list sessions, and process text data. It uses these tools to interact with local directories such as ~/.claude/projects/ and ~/.claude/claude-code-token-saver-data/.
[COMMAND_EXECUTION]: The skill employs python3 with heredoc strings in SKILL.md (Step 5a, 5c, 5d) to execute dynamically generated scripts for parsing JSONL transcripts and assembling restored context files. This script generation is used for local data processing but represents a dynamic execution pattern.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it restores content from previous sessions into the current context without sanitization.
Ingestion points: Reads session data from JSONL files in ~/.claude/projects/ and preprocessed text in ~/.claude/claude-code-token-saver-data/.
Boundary markers: Absent. The skill appends restored context to the conversation without specialized delimiters or warnings to ignore embedded instructions.
Capability inventory: Executes shell commands (node, python3, git, sed, awk), reads/writes local files, and performs git operations.
Sanitization: Absent. While preprocess.js truncates messages and merges tool calls, it does not filter or escape potentially malicious instructions embedded in the historical messages.

continue