Plugin Claude Code
Wires the AURORA Agent (bioprism) MCP server into Claude Code: 259 tools for decision-context compilation (FIBER), missions, evaluation, capability routing, and the autonomous-brain control plane. Requires a built aurora-agent checkout (AURORAAGENTROOT or /aurora-agent).
Skill Claude CodeCodex
Set up, build, or troubleshoot the AURORA Agent (bioprism) MCP server used by the aurora-backend plugin. Use when the aurora-agent MCP server fails to start, when its tools are missing from a session, when asked to install or build the aurora-agent backend, or when a bioprism build or test behaves strangely on Windows.
Plugin Claude Code
Drive the FIBER decision-context compiler from any project: compile, explain, compare against baselines, validate worlds, and verify certificates, with the 10-exit-code retryability matrix interpreted for you. Requires a built aurora-agent checkout.
Plugin Claude Code
The honest-labelling engineering discipline from the AURORA Agent workspace, packaged for any codebase: never collapse an error into a benign default, make invariants unrepresentable, prove your scanners fire, and recognise Windows test-runner lies.
Skill Claude CodeCodex
Apply the honest-labelling discipline when designing APIs, reporting results, writing benchmarks, or reviewing code that publishes numbers or claims. Use when a value could mean either "measured as zero" or "never measured", when a success flag could hide an unchecked path, when reporting benchmark or evaluation…
Skill Claude CodeCodex
Diagnose Windows test runs that lie — os error 4551 means the binary never executed, and a workspace test total can silently lose every suite after a blocked one. Use when cargo test reports a failure with no failing test named, when a test count looks lower than expected on Windows, or when a freshly built binary…
Plugin Claude Code
Integrate applications with the aurora-agent backend: launch the bioprism-api HTTP gateway, use the dependency-free TypeScript and Python SDKs, and test MCP wiring without the Rust binary.
Plugin Claude Code
Author and run aurora-agent missions — validated DAGs of MCP tool calls with least-authority policies, preflight that dispatches nothing, and fail-closed operator gate acceptance. Requires the aurora-backend plugin (or the bioprism-api gateway).
Plugin Claude Code
Operations evidence for the aurora-agent control plane: an ops-auditor subagent and a health command that read snapshots, capability gates, and the recovery matrix — reporting evidence postures verbatim, never collapsing them into readiness.
Plugin Claude Code
The measurement-science methodology from the AURORA Agent workspace, packaged for any evaluation or benchmarking effort: equal-engineering baselines, discriminating experiment design, certificate-audited analysis, metamorphic evaluation, honest figures, and research-dossier discipline.
Skill Claude CodeCodex
Treat every computed result as unverified until its certificate checks out, review what was omitted rather than only what was included, and digest-chain every published number back to a retained computation. Use when citing any compiled or pipeline output in a document, when reviewing an analysis that publishes…
Skill Claude CodeCodex
Recognize when a benchmark cannot discriminate between methods, construct experiment families that could, sweep structural knobs without sweeping the decision itself, and report a tie that survives as the headline result. Use when all methods score the same on your benchmark, when designing a new evaluation world or…
Skill Claude CodeCodex
Build and tune baselines with the same engineering effort as the system under test, rank them on admissibility rather than on getting the right answer, and render refusals as refusals rather than zeros. Use when writing or reviewing any benchmark or comparison table, when a baseline looks suspiciously weak, when a…
Skill Claude CodeCodex
Make every figure and table verifiable and non-misleading - source digests attached, refusals and absences drawn as such rather than as zeros, ties and negative results at equal visual prominence, no truncated-axis emphasis, and captions that state what the figure cannot show. Use when producing any figure, table…
Skill Claude CodeCodex
Grow an evaluation suite by mutation with executable postconditions, count effective diversity in equivalence classes rather than instances, and report yield honestly. Use when generating benchmark variants or augmented test cases, when a suite's instance count is quoted as evidence of coverage, when checking whether…
Skill Claude CodeCodex
Scope what an autonomous research run may claim and what it must never claim, carry limitations verbatim through every summary layer, and treat negative findings as first-class results. Use when writing up results from an autonomous or agent-driven measurement run, when drafting a research summary, README section, or…
Plugin Claude Code
Engineering skills for working inside the aurora-agent (bioprism) workspace: adding crates, byte-parity checks, crate verification, blueprint classification, and boilerplate measurement. Only useful with the aurora-agent repo checked out.