The Agent Skills Directory

[DATA_EXFILTRATION]: The skill is designed to access and read files within ~/.claude/projects/, which contains the full JSONL transcripts of a user's interactions with Claude Code. These transcripts are highly sensitive as they contain the entire conversation history, including code snippets, tool outputs, and any secrets or private information shared during the session. While the skill primarily extracts token counts and timestamps, it also generates a summary of this sensitive content and transmits it to an external Tana workspace.
[COMMAND_EXECUTION]: The skill executes multiple shell commands and a local script to gather context. It runs git diff --stat and git log to inspect repository history, and it invokes python3 scripts/session-stats.py to perform file system discovery and parsing of the sensitive transcript files.
[INDIRECT_PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by ingesting and processing untrusted data from the local environment.
Ingestion points: The skill reads git diff output, git log entries, and session transcripts from ~/.claude/projects/.
Boundary markers: There are no explicit delimiters or "ignore instructions" warnings defined when the agent processes these inputs to generate a summary.
Capability inventory: The skill possesses the capability to execute shell commands (Bash), run Python scripts, and send data to an external service via the import_tana_paste tool.
Sanitization: No sanitization or filtering is performed on the ingested git data or transcript content before it is processed by the agent to create the Tana entry.

tana-session-log