The Agent Skills Directory

[DATA_EXFILTRATION]: The skill indexes local interaction history from ~/.claude/projects and ~/.codex/sessions. If a remote central service is configured via the RECALL_URL environment variable or a configuration file, the skill can synchronize session data and search results to an external network endpoint.
[PROMPT_INJECTION]: Because the skill retrieves and presents content from previous AI conversation transcripts, there is a risk of indirect prompt injection. Malicious instructions embedded in historical records could influence the agent's behavior when those records are retrieved.
Ingestion points: Local JSONL transcript files located in Claude and Codex session directories, processed by scripts/recall.py.
Boundary markers: The skill does not explicitly instruct the agent to use delimiters or "ignore instructions" wrappers when processing retrieved transcript content.
Capability inventory: The core scripts/recall.py utility has the ability to read files, manage a local SQLite database, and perform network requests to a central service.
Sanitization: The utility implements robust regular expression filtering (e.g., SECRET_RE, PRIVATE_KEY_RE) to automatically identify and redact API keys, private keys, and other credentials from session data before it is indexed or transmitted.
[COMMAND_EXECUTION]: The skill instructions guide the agent to perform searches and manage the index using shell commands, including python3 scripts/recall.py and standard utilities like rg and grep.

recall