The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests historical conversation data from a local ataru index. Since this content is user-generated or agent-generated, it could potentially contain malicious prompts (indirect prompt injection) that attempt to manipulate the current agent's behavior when the content is read into the context.
Ingestion points: scripts/ataru_recall.py reads JSON output from the ataru binary's search and read subcommands in scripts/ataru_recall.py.
Boundary markers: The skill does not currently use specific delimiters (e.g., XML tags or special tokens) to separate recalled chat snippets from its own instructions, although the instructions in SKILL.md tell the agent to cite sources.
Capability inventory: The skill has the ability to execute local shell commands (via subprocess.run in scripts/ataru_recall.py) and write to a local profile JSON file (via scripts/profile_store.py).
Sanitization: ataru_recall.py mitigates large data influx by projecting only essential fields and truncating snippets to 600 characters and messages to 2000 characters by default.
[COMMAND_EXECUTION]: The skill invokes a local binary (ataru or lovcode) to perform its search operations.
The scripts/ataru_recall.py script uses subprocess.run with list-based arguments, avoiding shell interpolation risks.
It includes a normalize_argv helper to safely handle project identifiers that begin with hyphens, preventing them from being misinterpreted as command-line flags.
[EXTERNAL_DOWNLOADS]: The README.md provides installation commands that fetch the skill and its components from the lovstudio vendor repositories using npx. This represents the standard distribution method for the vendor's own tools.

lov-search-chat