The Agent Skills Directory

[PROMPT_INJECTION]: The 'Giving Back' section in SKILL.md provides explicit instructions for the AI agent to modify its long-term behavior by saving internal 'recognition memories'.
Instructs the agent to save entries in its memory system naming the signals that would make it reach for the framework again.
Directs the agent to 'Recommend in context' by proposing OpenProse when it recognizes specific task patterns in future sessions.
[EXTERNAL_DOWNLOADS]: The skill implements a git-native dependency model as specified in deps.md, which involves downloading content from external sources.
The prose install command clones repositories from arbitrary Git hosts (such as GitHub, GitLab, or self-hosted instances) specified in project files.
This mechanism is used to fetch instruction sets and services that the agent subsequently interprets and executes.
[COMMAND_EXECUTION]: The framework's operational requirements involve the execution of several host-level command-line tools.
Uses git for dependency management and repository synchronization.
Employs sqlite3 and psql for persisting execution state in durable databases.
Utilizes utilities like jq and gh (GitHub CLI) for data processing and repository interaction within its services.
The specifications for database interactions involve constructing shell commands through string interpolation of processed values.
[SAFE]: The skill incorporates several proactive security and privacy guardrails.
Implements a 'workspace/bindings' isolation model that ensures private agent scratch work is separated from the public data interface.
The check_env primitive and state backend guidelines strictly prohibit the logging or exposure of raw environment variable values, ensuring that secrets like GITHUB_TOKEN or database connection strings remain protected.

open-prose