Indium-AI-Labs

Indium-AI-Labs/indium-agentkit

Skill Codex

Review Dockerfiles, Terraform and other infrastructure-as-code, Kubernetes and Helm manifests, CI deploy paths, and cloud configuration for security, reliability, and cost risks, producing a verified findings register. Use when infrastructure changes need review before merge or an environment needs a posture check.…

Indium-AI-Labs/indium-agentkit

Skill Claude CodeCodex

Design and run evaluations for prompts, models, RAG pipelines, and agents: success criteria tied to the product, datasets from real and adversarial cases, code, model, and human graders, calibrated judges, repeated trials with confidence intervals, baseline gating, and cost and latency. Use when a prompt, model…

Indium-AI-Labs/indium-agentkit

Skill Claude CodeCodex

Design and run load, stress, spike, and soak tests with k6, Locust, or the project's tool against an approved non-production target, using open-model arrival rates, SLO-based thresholds, and resource monitoring to find capacity limits. Use when validating throughput or latency targets, sizing capacity, or checking…

Indium-AI-Labs/indium-agentkit

Skill Claude CodeCodex

Run an open-weight model locally with Ollama, llama.cpp llama-server, or vLLM behind an OpenAI-compatible endpoint: estimate weight and KV-cache memory first, pick quantization and engine, bind to loopback, health-check, and measure streaming time to first token. Use when adding, sizing, or debugging a local or…

Indium-AI-Labs/indium-agentkit

Skill Claude CodeCodex

Prepare, audit, and stage an iOS or Android release (native, React Native, Expo, or Flutter): version and build numbers, signing and secrets, permissions and privacy manifests, store policy, symbol upload, phased or staged rollout with halt criteria, OTA update scope, and kill switches. Use before submitting to App…

Indium-AI-Labs/indium-agentkit

Skill Claude CodeCodex

Add or improve structured logging, metrics, OpenTelemetry tracing, and alerts in application code using the project's existing libraries, with central redaction, bounded label cardinality, and actionable SLO-based alerts that link a runbook. Use when a service cannot answer an operator question from its signals, new…

Indium-AI-Labs/indium-agentkit

Skill Claude CodeCodex

Produce a verified onboarding guide for a repository covering purpose, prerequisites, setup, source layout, key abstractions, common workflows, and gotchas, with every setup and test command run or marked unverified. Use when new contributors need a persistent guide to get from clone to a passing test run and their…

Indium-AI-Labs/indium-agentkit

Skill Claude CodeCodex

Speed up a specific slow path (latency, throughput, CPU, memory, or startup) by measuring a reproducible baseline, profiling to find the bottleneck, making the smallest change, and proving the gain against a separate baseline build with a statistical comparison. Use when something is measurably slow or has regressed…

Indium-AI-Labs/indium-agentkit

Skill Codex

Turn a feature request, bug report, refactor, or technical proposal into an implementation-ready plan with goals, non-goals, acceptance criteria tied to tests, affected code, ordered reviewable steps, risks, and rollback. Use when a change is non-trivial, spans several files, or needs agreement before coding. Not for…

Indium-AI-Labs/indium-agentkit

Agent Claude Code

Investigates a bug, failing test, or regression by reproducing it, tracing the divergence, and testing competing hypotheses, then returns the root cause with evidence and no fix. Use when the cause of wrong behavior is unknown and evidence gathering can run in parallel with other work. Read-only.

Indium-AI-Labs/indium-agentkit

Agent Claude Code

Researches authoritative external documentation for the exact library, framework, SDK, or API versions a project uses, and returns version-tagged, cited facts to prevent hallucinated APIs. Use when code depends on an API whose signature, behavior, configuration, or version differences are uncertain, or before an…

Indium-AI-Labs/indium-agentkit

Agent Claude Code

Reviews frontend components, state management, rendering boundaries, styling tokens, and bundle impact for correctness and performance risks. Use proactively after UI changes or before merging frontend work. Not for accessibility audits (use the accessibility-checker subagent). Read-only.

Indium-AI-Labs/indium-agentkit

Agent Claude Code

Reviews Dockerfiles, Terraform and other infrastructure-as-code, Kubernetes and Helm manifests, and cloud configuration for security, reliability, and cost risks. Use proactively before merging infrastructure changes or when an environment's posture is in question. Read-only.

Indium-AI-Labs/indium-agentkit

Agent Claude Code

Critiques an implementation plan against the actual codebase for missing steps, wrong assumptions, unhandled risks, scope creep, and unverifiable acceptance criteria, and proposes concrete amendments. Use before executing a non-trivial plan, or when a plan touches unfamiliar code, public contracts, data, or…

Indium-AI-Labs/indium-agentkit

Agent Claude Code

Adds or repairs automated tests at public behavior seams for assigned behavior, including characterization tests before a refactor and regression tests for a fixed bug, and proves they fail and pass for the right reasons. Use when test coverage is missing or a change needs red-green evidence. Edits only assigned paths.

Indium-AI-Labs/indium-agentkit

Skill Claude CodeCodex

Respond to pull request review comments: fetch every review thread, classify each comment as must-fix, agree-and-fix, question, disagree, or out of scope, make minimal focused changes, run the checks, and draft replies that reference each change. Use when a PR has review comments or requested changes, or the user asks…

Indium-AI-Labs/indium-agentkit

Skill Claude CodeCodex

Implement a product feature backed by a hosted large language model API: task contract, prompt design, structured outputs, tool use, streaming, timeouts and retries, caching, cost controls, prompt-injection defenses, privacy, and evaluation before release. Use when adding or changing code that calls an LLM provider…

Indium-AI-Labs/indium-agentkit

Skill Claude CodeCodex

Author or improve a Dockerfile, .dockerignore, and Compose setup for a service with multi-stage builds, digest-pinned base images, a non-root user, signal-safe startup, health checks, no secrets in layers, and image scanning plus an SBOM. Use when containerizing a service, slimming or hardening an image, or setting up…

Indium-AI-Labs/indium-agentkit

Skill Claude CodeCodex

Diagnose and fix tests that pass and fail on the same code: measure the flake rate, classify the cause (timing, ordering, shared state, environment, concurrency, external dependencies), fix the root cause, and quarantine only with a tracked issue. Use when a test fails intermittently locally or in CI, or a retry makes…

Indium-AI-Labs/indium-agentkit

Skill Claude CodeCodex

Create or refresh a project's AGENTS.md, and CLAUDE.md for Claude Code, from repository evidence: what the project is, verified test and lint commands, source and test locations, conventions, and agent gotchas, kept short because it loads into every session. Use when a repository has no agent context file, the…

Indium-AI-Labs/indium-agentkit

Skill Claude CodeCodex

Build or review a Model Context Protocol (MCP) server that exposes tools, resources, or prompts to AI clients: SDK and transport choice, model-facing tool names, descriptions, and schemas, tool errors, authorization, least privilege, injection defenses, testing with the MCP Inspector, and backward-compatible…

Indium-AI-Labs/indium-agentkit

Skill Claude CodeCodex

Design LLM workflows that use more than one model call or agent: pick among prompt chaining, routing, parallel fan-out, orchestrator-workers, and evaluator-optimizer loops, then define termination bounds, handoff state, per-role tool permissions, human checkpoints, retries, tracing, and evals. Use when planning or…

Indium-AI-Labs/indium-agentkit

Skill Claude CodeCodex

Answer one technical question with a small, time-boxed, throwaway experiment that produces measured evidence and a go, conditional-go, or no-go recommendation before committing to an approach. Use when feasibility, performance, or integration behavior of a library, service, or design is uncertain. Not for production…

Indium-AI-Labs/indium-agentkit

Skill Claude CodeCodex

Diagnose and fix an existing slow database query, PostgreSQL first with MySQL notes, by capturing its real parameters, reading EXPLAIN ANALYZE plans, and verifying a targeted index, rewrite, or statistics fix. Use when a known query, endpoint, or job is slow, a plan regressed, or N+1 queries or deep OFFSET pagination…