human-in-the-loop plugins

by Databricks

Consort: a Spec-First Branched-Database TDD agent framework for building transactional applications on Lakebase (Databricks' serverless Postgres/OLTP database, not the Delta Lakehouse). A deterministic orchestrator drives role agents through a spec-first design lane and a test-driven build lane whose every cycle runs against a live copy-on-write branch of a real Postgres database, with human-in-the-loop gates that fail closed. For application backends on Postgres, not ETL or analytics. Run /consort:start to launch the workflow.

by Diego Romero

Autopilot for spec-driven development. The human signs off and merges; planwright flies the spec. Scales to a multi-spec fleet: an agentic software factory on Claude Code primitives.

by Halyard Labs

Shared team memory for AI agents. Log decisions, recall past work, gather context, and ask teammates on Slack, Teams, or Discord when you need a human.

by Junjie Luo

In-lab human interaction with deployed prediction endpoints — the Claude Code chat is the UI. CONSOLE mode (default): patient-first on-demand inference — list patients, view the full chart, list available models (Endpoint_Set packages + live status), and run 'this model on this patient': the endpoint-predict MCP tool prepares the payload (required tables + trigger record), POSTs to the endpoint, and returns the score verbatim with a data-gaps report. STUDY mode: formal reader protocol (frozen review_bundle.json, blind→assisted session, decision-influence metrics vs gold). Endpoint-agnostic: local Flask, Databricks, SageMaker — same wire contract.

by Ciprian Ursu

Pre-execution risk gate and prompt injection guard for Codex CLI: judges every Bash command and subagent spawn before it runs, with local rules, an editable policy, optional TypeSafe Jev (System One) judgments, System 2 escalation and a human approval queue. Shadow mode by default, no API key needed.

by Jordan Richlen

Classify a candidate action as PATCH/MINOR/MAJOR (semver-style blast-radius test) before acting — act silently on PATCH, flag-and-stage MINOR, stop for explicit human sign-off on MAJOR. Use whenever you're mid-task and unsure how much autonomy to take on the next action: which of several implementation paths to pick, whether to overwrite unreviewed state, whether to disable a safety toggle, or any judgment call settings.json's autoMode patterns don't enumerate.

by Junjie Luo

Human-grounded subjective labeling in two sibling sides: Label Building calibrates and freezes one identified human's construct into a signed Label Handoff; Label Scanning qualifies executors on sealed human gold, labels the corpus under that frozen handoff, and closes with an independent probability audit.

by Rugby Waldorf LLC

Write a document and its flowchart with a human: open both in a browser, block while they edit, and get their revisions and comments back as JSON.

Real-time, advice-only document review for human-in-the-loop LLM sessions.

by Pushary

Push notifications, human-in-the-loop questions, and permission gating for your AI agent. Get a push when a task finishes, answer the agent from your phone, and approve risky commands before they run.