Agent Cards v0.1 draft. Declarative JSON document disclosing an AI agent's capability surface, refusal behavior, evaluations, and deployment posture. Like HuggingFace model cards, but for agents. Part of the Kinetic Gain Protocol Suite.

Fleet-level conformance scanner over a directory of Kinetic Gain Suite JSON docs. Sniffs the protocol of every file, counts by spec, validates required top-level blocks, reports unknowns and malformed files. Library + CLI.

Renders a Markdown README from an A2A AgentCard JSON via agent-card-readme-generator. Posts as PR comment AND/OR writes to a target file path for auto-synced docs. First in the readme-generator Action quintet.

Fleet-analyze a directory of A2A AgentCard documents. Counts by autonomy_level / memory_persistence, flags autonomous-without-IRU, destructive-on-non-autonomous, persistent-memory-without-refusal-taxonomy. Library + CLI.

Curated conformance corpus for agent-cards-spec v0.1 — categorized JSON fixtures (valid / dirty / shape) with declarative manifest. Final piece of the test-vectors corpus pattern across all 4 Kinetic Gain Suite specs.

GitHub Action wrapping agent-card-fleet-summary. Walks a dir of AgentCards, surfaces governance gaps (autonomous-without-IRU, destructive-on-non-autonomous, etc.), posts Markdown PR comment, fails on high findings. Sibling of llm-cost-rollup-action.

Diff two A2A AgentCard documents — classify changes (autonomy escalation, memory escalation, tool/model removal, incident-response URI removal, etc.), surface a breaking flag. The agent-cards counterpart to mcp-tools-diff. Library + CLI.

Generate a human-readable Markdown README from an A2A AgentCard JSON. Documents identity, capabilities, models, tools (with side-effect-class badges), refusal taxonomy, evaluations, deployment, safety posture. Sibling of mcp-server-readme-generator for the A2A side.

Sniff which protocol a JSON document represents — agent-cards-spec, mcp-tool-card-spec, prompt-provenance-spec, evidence-bundle-spec, OTel GenAI OTLP envelope, or MCP tools/list. Routing primitive in front of any governance pipeline.

Open 30-attack prompt-injection corpus + Python harness with Agent Card refusal-taxonomy back-references. 10 attack categories, 4-level severity, 5 scoring methods. Pairs with the Kinetic Gain Protocol Suite.

Visual web harness for prompt-injection-bench. Paste a JSONL transcript, see a procurement-grade pass-rate report. Live at bench.kineticgain.com. Client-side React + TS.

Conformance check between an A2A AgentCard's declared tools and an MCP server's tools/list. Reports missing (declared but not exposed), undeclared (exposed but not in card), matched; counts by side-effect class. Library + CLI.

Build + validate a minimal A2A AgentCard per agent-cards-spec v0.1 from a StampOptions struct. Enforces id/version regex, enum constraints, and the spec's autonomy→IRU allOf clause. Sibling of prompt-provenance-stamp.

Unified MCP server for all 11 Kinetic Gain Protocol Suite specs + DefenseTech module (8 tools) — 71 callable tools (47 spec + 16 implementation + 8 DefenseTech). One Claude Desktop config entry. Stdio, no API key, no build step.