The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The README.md provides a one-line command to install the skill pack using curl to fetch a script from raw.githubusercontent.com/parhumm/jaan-to. As this is the official repository of the skill author, it is categorized as a safe vendor resource.
[REMOTE_CODE_EXECUTION]: The installer in README.md executes the remote installation script via bash process substitution. This is the standard installation method for this vendor's distribution.
[INDIRECT_PROMPT_INJECTION]: Several skills ingested untrusted data from repository files, pull request diffs, or external issues. The framework addresses this risk by providing an explicit defense architecture documented in the security guardrails blueprint, including model safety instructions, input envelopes, and role separation between planning and execution agents.
[COMMAND_EXECUTION]: The skill executes various system tools like git, gh (GitHub CLI), and curl. These operations are strictly scoped in the allowed-tools metadata and are monitored by a security gate script (pre-tool-security-gate.sh) that blocks unauthorized patterns such as sudo or unsafe pipes to shell.
[PROMPT_INJECTION]: Heuristic detections of prompt injection patterns in documentation files (e.g., runtime/docs/research/79-jaanto-security-guardrails-blueprint.md) are benign, as these files describe potential threats and defensive strategies for training and detection purposes.

jaan-to-codex-pack