Return ``(available, reason)``: the binary is on PATH AND authenticated.

Quick connectivity check against the OSV.dev API (HEAD-like POST).

Binary on PATH AND authenticated (delegates to the registry probe).

Total number of keys in the pool.

Return ``context_length`` for *model* from the resolved YAML registry.

Return ``max_output_tokens`` for *model* from the resolved YAML registry.

SkillSpector - Security scanner for AI agent skills (LangGraph). Analyze skill bundles to detect vulnerabilities and security risks. Sup

Create a deterministic dummy finding for a stub analyzer.

Create a malicious skill directory for testing.

Run SkillSpector as an MCP server. Exposes a single tool, ``scan_skill``, so any MCP-capable agent (Claude Code, Codex CLI, Gemini CLI)

Patch get_chat_model; configure response via mock_llm.ainvoke = AsyncMock(...).

Patch the layered resolver so tests never make network requests.

Create a directory with 3 sub-skills, no root SKILL.md.

A directory with root SKILL.md AND sub-skill SKILL.md files.

Analyze MCP tool manifest for tool-poisoning indicators (TP1-TP4).

Run prompt_injection patterns and return findings.

Run data_exfiltration patterns and return findings.

Parse Python files and detect source\u2192sink data flows.

Run SSRF patterns and return findings.

Run excessive_agency patterns and return findings.

Run rogue_agent patterns and return findings.

Run anti_refusal patterns and return findings.

Analyze manifest permissions vs code capabilities; emit LP1-LP4 findings.

Run output_handling patterns and return findings.

Run tool_misuse patterns and return findings.

Run privilege_escalation patterns and return findings.

Run memory_poisoning patterns and return findings.

Run supply_chain patterns (SC1–SC6) and trigger analysis (TR1–TR3).

Run agent_snooping patterns and return findings.

Run harmful_content patterns and return findings.

Run system_prompt_leakage patterns and return findings.

Parse Python files via AST and detect dangerous execution patterns.

Organize files in the given directory by extension.

Read-only: parses records and returns a list. Makes no changes to disk.

Register custom markers for the contrib.multilingual test suite.

Total number of 429 responses encountered across all keys.

Read a config file and return its contents.

No HTTP credentials needed — the CLI handles auth itself.

Bedrock uses SigV4, not ``(api_key, base_url)`` — always returns ``None``. ``providers.resolve_chat_model_credentials`` treats ``None`` as "t

Return the model to forward to the CLI. CLI providers default to the user's OWN CLI-configured model (we omit ``--model`` entirely),

Resolve model: ``SKILLSPECTOR_MODEL`` env > slot default > DEFAULT_MODEL.

Resolve model: ``SKILLSPECTOR_MODEL`` env > slot default > ``DEFAULT_MODEL``.

Resolve model: ``SKILLSPECTOR_MODEL`` env > slot default > ``DEFAULT_MODEL``.

Resolve model: ``SKILLSPECTOR_MODEL`` env > slot default > ``DEFAULT_MODEL``.

Total number of successful retries after a key switch.

Run the MCP server over ``stdio`` (local agents) or ``http`` (remote/A2A).

Create a safe skill directory for testing.

Scan a skill for security vulnerabilities. Examples: skillspector scan ./my-skill/ skillspector scan ./my-skill/ --format j

Scan an AI agent skill for security risks before installing it. Use this before installing or loading any skill or MCP server to decide

#F1: One shared analyzer for all tests — avoids repeated ChatOpenAI creation.

Create a single-skill directory with root SKILL.md.

All bytes written to stdin by the bounded reader.

Summarize text by calling a remote API.

Restore global state mutated by setup_deepseek_compat(). Calls _restore_patches until depth reaches 0 (setup may be called multiple ti