The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches prebuilt native binaries and release manifests from the vendor's repository.\n
Evidence: scripts/setup_native.py downloads archives from github.com/EdukeyTeam/agent-toolbox/releases.\n
Evidence: scripts/context7_backend.py downloads model weights from Hugging Face for semantic search.\n- [REMOTE_CODE_EXECUTION]: Automates the download and execution of native binaries with built-in integrity verification.\n
Evidence: scripts/setup_native.py verifies SHA-256 hashes of downloaded ZIP and tar.gz archives and individual member files against a signed release manifest.\n
Evidence: The script performs a smoke test by executing the binary with a fixture repository after installation.\n- [COMMAND_EXECUTION]: Executes system utilities and local helpers to perform codebase analysis.\n
Evidence: scripts/repo_files.py executes git commands to inventory files and check ignore status.\n
Evidence: scripts/context7_backend.py spawns node to run JavaScript helpers for inference tasks.\n- [DYNAMIC_EXECUTION]: Implements dynamic loading for internal scripts and external libraries.\n
Evidence: scripts/legacy_tools.py uses importlib.import_module to load command handlers.\n
Evidence: scripts/retrieval/embed.mjs uses import() to load the @huggingface/transformers package from a user-defined runtime path.\n- [INDIRECT_PROMPT_INJECTION]: Processes untrusted repository content through a retrieval system.\n
Ingestion: scripts/context7_backend.py reads code and documentation from local repositories.\n
Boundary Markers: Retrieval payloads are structured in JSON with provenance metadata (kind, path, revision, sha256).\n
Capability Inventory: The skill has file-write (SQLite index), network-read (model downloads), and subprocess-execution capabilities.\n
Sanitization: scripts/repo_render.py sanitizes non-printable characters in rendered snippets to prevent terminal or display confusion.

legacy-codebase-workflows