The Agent Skills Directory

[PERSISTENCE]: The re0-upgrade skill (skills/breadth/re0-upgrade/SKILL.md) automates the configuration of persistent behavior by wiring a SessionStart command into agent configuration files such as ~/.claude/settings.json and ~/.codex/config.toml. This command is configured to execute a downloaded script located at ~/.re0/session-check.cjs every time a new session begins.
[EXTERNAL_DOWNLOADS]: The re0-upgrade skill downloads external JavaScript files, including catalog.cjs, session-check.cjs, and opencode-discovery.js, from the vendor's GitHub repository at raw.githubusercontent.com/LilMGenius/paperthin to the user's local directory ~/.re0/.
[REMOTE_CODE_EXECUTION]: By downloading external scripts and configuring them to run automatically at session startup, the skill implements a remote code execution pattern. This allow logic sourced from a remote repository to be executed in subsequent agent sessions without further user interaction.
[COMMAND_EXECUTION]: Multiple skills, including re0-release, re0-git, and re0-upgrade, facilitate the execution of shell commands such as git, npm, npx, and gh. Specifically, re0-upgrade utilizes the GitHub CLI (gh api) to perform authenticated API calls on the user's account to star the repository.
[INDIRECT_PROMPT_INJECTION]: The skill suite ingests untrusted external artifacts through the engineering_artifact input, creating a surface for indirect prompt injection attacks.
Ingestion points: The engineering_artifact input parameter is used across several skills, including fable-wise and autobahn.
Boundary markers: While the autobahn skill (skills/depth/autobahn/SKILL.md) mentions 'FRAME', 'CARVE', and 'GUARD' steps for defining scope, it does not implement deterministic delimiters for the external content.
Capability inventory: The skill possesses capabilities for file system access, shell command execution, and network access through the update and session-check scripts.
Sanitization: The skill relies on LLM-based reasoning in its 'CARVE' and 'GUARD' phases to filter risky content, which represents a probabilistic defense rather than a strict sanitization protocol.
[TIME_DELAYED_CONDITIONAL]: The session-check.cjs script implements a time-based gate using a THROTTLE_MS variable, which ensures that the discovery notice logic only executes once every 24 hours.

fable-wise