The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from the Reticulum mesh network, creating a surface for indirect prompt injection.
Ingestion points: Reads discovery announces, LXMF contact addresses, and network packet data (SKILL.md, references/discovery.md).
Boundary markers: The instructions emphasize redacting keys and keeping responses small, but lack explicit delimiter markers or instructions to ignore commands embedded in mesh data.
Capability inventory: The skill can perform file transfers (rncp), remote execution (rnx, rnsh), and interact with the local daemon (rnsd) (SKILL.md).
Sanitization: Instructions mandate the redaction of private keys and passphrases before surfacing data to the user.
[DATA_EXFILTRATION]: The skill accesses sensitive local configuration paths.
Sensitive paths: Accesses ~/.reticulum/config, ~/.reticulum/storage, and ~/.reticulum/identities (SKILL.md). These files contain interface credentials, passphrases, and cryptographic identities.
Mitigation: The skill includes explicit instructions for the agent to redact passphrases, keys, and credentials before outputting data.
[COMMAND_EXECUTION]: The skill utilizes a wide array of command-line utilities for protocol operations.
Tools: Includes rnsd, rnstatus, rnpath, rnprobe, rncp, rnid, rnx, rnsh, and rnodeconf.
Remote Execution: Specifically notes that rnx and rnsh allow for remote execution, warning the user about 'no-auth' modes (SKILL.md).
[EXTERNAL_DOWNLOADS]: The skill references the installation of Reticulum ecosystem packages from public registries and mesh-based remotes.
Packages: Mentions pip install for rns, lxmf, nomadnet, and lxst via PyPI (SKILL.md).
RNS Remotes: Describes pip-rns, which installs software directly from Reticulum rngit remotes (rns:// paths) (SKILL.md). These are standard ecosystem components but involve downloading code from decentralized sources.

reticulum