alankatanoisi

anthropic-official

Anthropic policy and product posture for the playground harness—official citations (June 15 Agent SDK credits, OAuth vs API keys, claude -p vs API), surface picker (API vs Agent SDK vs Claude Code), and matrix seed rows. Use when Alan asks what Anthropic documents, which integration surface to use, or to refresh lab-notes/parity/anthropic-official-posture.md.

lab-integrator

Keeps playground lab-notes aligned—weekly-integration.md rollups, cross-links between parity/observability/oauth docs, agents README freshness. Use when Alan asks for weekly lab summary, doc index cleanup, or integration note after multiple lanes landed files.

Code & Development Listed

oauth-evidence

Writes OAuth-only headless demo runbook and bench parity evidence templates for the playground—debug preflight, golden runner commands, artifact checklist, known cache-TTL blocker. Use when Alan asks how to demonstrate Pro/Max OAuth via bridge without API keys; do not run live calls unless Alan explicitly asks.

observability-scribe

Documents runner observability contract—stream-json event types, stopReason taxonomy, usage fields, autopsy/ledger shapes—from kernel contract and tests. Use when Alan asks for observability-contract.md, stream-json catalog, or stop reason inventory.

parity-archivist

Maintains Claude Code parity lab-notes—claude-parity-matrix.md (adopt/skip/later + evidence), permission-modes.md, structured-output.md—from HARNESS_VISION and runner code. Use when Alan asks for parity matrix, permission mode mapping, or structured output design vs runner today.

runner-command-builder

OUTPUT TERMINAL COMMANDS ONLY — never edit repo files in this chat. When Alan says runner-command-builder, commands only, or paste runner CLI: return three bash blocks (A as-asked, B safer, C performance) with one-line rationale and tip each — nothing else. Do NOT use anthropic-platform-expert, parity-archivist, or any lab-notes skill. Words implement/fix/update go inside the quoted node bin/local-bridge-runner.js prompt, not Cursor edits.

anthropic-platform-expert

Anthropic platform expert — Agents SDK, Claude API, official docs, changelogs, and public status (X) for bridge/runner parity.