stop-hook

Gemini stands at the door. Independent quality gate that audits Claude Code's output before it can stop.

Claude Code Stop hook that nudges granular commits after every turn. Guarded for rebases, merges, detached HEAD. Pure bash, zero deps.

The creator kit: 139 Claude Code commands, 22 skills and 5 agents that carry their own DTD grammar, laws and trust boundary. Creators for prompts, skills, hooks, commands, subagents, plans, MCP servers, workflows, tasks and filetypes. The Adiutor checks every answer against the DOCTYPE that produced it; the version is recognised, not typed.

Your agent said Done. nuhuh runs the experiment with fresh tests, real exit codes and actual files, then hands back a receipt it wrote, not one your agent dictated.

Auto-recover Claude Code when Opus tool calls fail to parse ('retry also failed'). A Stop hook that detects the terminal failure and re-prompts the model to retry with a shorter call.

A /goal engine for Claude Code where completion is EARNED, not announced: acceptance criteria are real shell commands, the Stop hook re-runs every one of them, and only exit 0 ends the session.

Claude Code plugin: queue prompts with /queue that fire at end-of-turn, after Claude fully finishes — never mid-task. A FIFO prompt queue via a self-terminating Stop hook. Installable as a marketplace.

cc-vellum — digital vellum for Claude Code conversations. A Stop hook that durably archives every turn and generates AI summaries. Zero dependencies, pluggable providers (Claude CLI, OpenAI, Ollama).

Give Claude Code a voice on Windows: a Stop-hook + edge-tts daemon speaks each reply aloud, with a skill to switch voices, mute, and interrupt.

Defensive /goal condition templates for Claude Code · 4-piece recipe · 6 patterns · 8 NEVER rules · 11-check linter · self-validating CI · MIT

"Done" ≠ done. A Claude Code skill + Stop hook that blocks completion claims ("fixed", "added", "all set", "wrapped up") and negative-existence claims until verified with a tool call — so the agent stops reporting from memory and starts citing evidence.

Stop hook that blocks Claude from finishing until quality checks pass. Detects rationalized incompleteness, stale learning logs, and low disk space. 200-line Python, stdlib only, deterministic.

Won't let your AI agent quit early — a Stop-hook guard for Claude Code & Codex that keeps it working until the goals are provably done.

Custom status line for Claude Code: model, output TPS, token/cost totals, git branch, context usage — plus an opt-in auto-continue loop and a pricing updater. Zero-dependency single-file Node.js.

Guardrails for AI agents. HORKOS, god of oaths, checks the receipts: no proof, no "done". Evidence-audit hooks for Claude Code, zero LLM calls, deterministic benchmarks.

Native macOS notifications for Claude Code stop events — 3-4 word AI summaries, click-to-focus your specific terminal window, Claude-branded icon. Ships as a plugin and an install script.

macOS banner + applause + click-to-jump for long-running terminal tasks (Claude Code, Codex CLI)

Stop hook that catches low-agency language (can't/cannot) in Claude's replies and forces a reframe via a bundled high-agency skill.

A Claude Code Stop hook that blocks 'lies of completion' — forces verification before session ends. Free sample from the Claude Code Productivity Hook Pack.

Cross-session messaging for Claude Code on native Windows, where the built-in feature (macOS/Linux/WSL2 only) does not ship. Solves: a message appears on the recipient's screen but never reaches the model. A Stop hook delivers at turn end; an async watcher wakes idle sessions with no human input. PowerShell only.