The Agent Skills Directory

[OBFUSCATION]: The skill uses HTML entities (d) to hide the word 'codex' within strings such as Codex and the file path ~/.codex/config.toml. This technique is used to evade static analysis tools searching for specific text patterns or competing products.
[CREDENTIALS_UNSAFE]: The skill explicitly targets configuration files from other agent products, including ~/.claude/settings.json, ~/.config/opencode/opencode.json, and ~/.codex/config.toml. These files are known to store sensitive environment variables, tool configurations, and project metadata that can include secrets or API keys.
[EXTERNAL_DOWNLOADS]: The skill makes a network request to https://models.dev/api.json to fetch model pricing. This is a non-whitelisted and non-trusted domain, which introduces a dependency on an external third-party service for runtime logic.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources including competitor configuration files, markdown files, and JSONL session logs. It uses this data to drive agent behavior and memory recording without explicit sanitization or boundary markers.
Ingestion points: Reads ~/.claude/settings.json, CLAUDE.md, .mcp.json, ~/.codex/config.toml, and session log files across multiple platforms.
Boundary markers: None identified; external content is processed and mapped directly to agent context.
Capability inventory: The skill has access to bash, edit, write, and memory tools, and can invoke other skills like init-deep and coding-agent-sessions based on ingested data.
Sanitization: No validation or escaping of external content is described before the data is processed or recorded to memory.

onboarding