The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The setup.sh script fetches the Bun installer from https://bun.sh/install and clones the episodic-memory repository from GitHub (github.com/obra/episodic-memory.git) to provide semantic search functionality.
[REMOTE_CODE_EXECUTION]: The setup.sh script installs Bun by piping its installation script from bun.sh directly to a shell. It also executes npm install and npm link within the cloned episodic-memory repository to install the tool and its dependencies.
[COMMAND_EXECUTION]: The skill executes local shell commands using bun and the episodic-memory CLI to build search indexes, perform queries, and sync embeddings. These operations are managed via the setup.sh and search.ts scripts.
[DATA_EXFILTRATION]: The skill reads sensitive Claude Code conversation history logs stored in ~/.claude/projects/. The index-sessions.ts script parses these files to extract metadata, such as user messages, file paths, and PR links, to populate a local index file at ~/.claude/session-index/index.json.
[PROMPT_INJECTION]: The skill's process of reading and summarizing past conversation histories presents a surface for indirect prompt injection if the logs contain malicious instructions.
Ingestion points: Reads .jsonl session files from the local filesystem (File: index-sessions.ts).
Boundary markers: No explicit markers or instructions to ignore embedded commands are used during indexing or search result synthesis.
Capability inventory: Executes shell commands and external search utilities via child_process.spawn (File: search.ts).
Sanitization: Relies on JSON parsing and regular expressions for data extraction.

find-claude