HomeTrang chủ › All hooksMọi hook
The enforcement layer, by tier — 66 registered hooksLớp cưỡng chế, theo hạng — 66 hook đã đăng ký

Every hookMọi hook

66 hooks

A hook is a small program the harness runs around a tool call. Its tier is a code constant (HOOK_CLASS) — config can toggle a hook on/off but never change its tier, and the fail posture follows the tier. The 83 files under harness/hooks/ hold 67 tiered hooks (grouped below) and 16 shared libraries with no tier. See the rig for how the gate uses them.Hook là một chương trình nhỏ harness chạy quanh mỗi lần gọi tool. Hạng của nó là hằng số trong code (HOOK_CLASS) — config bật/tắt được nhưng không đổi được hạng, và tư thế-khi-lỗi theo hạng. 83 tệp dưới harness/hooks/ gồm 67 hook có hạng (nhóm bên dưới) và 16 thư viện dùng chung, không có hạng. Xem the rig để biết gate dùng chúng thế nào.

compliance

ComplianceCưỡng chế (16)

fail-closed, exit 2 — blocks the tool and prints the fix when it trips.

bash_safety_guard

PreToolUse(Bash) compliance hook: destructive-command gate.

context_firewall

PreToolUse gate on a tool call's blast radius.

disabled_skill_router

PreToolUse(Skill) block-with-map for an off skill.

floor_bash_guard

PreToolUse(Bash) compliance floor for Bash-spelled writes.

gate_stage

PreToolUse(Bash) compliance hook: the stage gate.

hook_dispatch

ONE command per (event, matcher) group.

instruction_policy_rewrite

PreToolUse(Bash): run the command the project's adopted policy prefers (`prefer-command`), in place of the one written.

model_guard

PreToolUse(Agent|Task) gate: enforce the model-bound posture.

plannotator_wrapper_guard

PreToolUse compliance gate: force the annotate review through the wrapper.

privacy_read_guard

put reads of secret files in front of the human (compliance).

protected_ref_guard

PreToolUse(Bash) compliance hook: protect refs.

secrets_inject

PreToolUse compliance hook: substitute ${secret:NAME} in memory, or block the call.

simplify_gate

PreToolUse(Bash) gate: oversized unsimplified diffs.

skill_quality_gate

write-time SKILL.md content gate (compliance, fail-open on crash).

spawn_provenance_workflow_guard

PreToolUse(Workflow) gate: enforce the spawn-provenance budget against the Workflow tool's OWN declared fan-out width (Ask-3 Layer-2).

write_guard

tool-mediated config-edit gate (compliance, fail-closed).

nudge

NudgeNhắc (30)

advisory on stderr — reminds, never blocks; default OFF.

agent_behavior_linter

observes PostToolUse/PreCompact, reports at Stop.

auto_decision_review_nudge

Stop nudge for unreviewed must-review auto-decisions.

backlog_capture_nudge

advisory reminder to record deferred work through the backlog register instead of hand-editing prose (nudge class).

backlog_hygiene_nudge

advisory reminder to close stale open backlog items before a publish stage (nudge class).

cook_delegate_nudge

advisory backstop for cook's per-phase posture.

cook_isolation_nudge

advisory /clear-before-cook reminder (nudge class).

decision_capture_nudge

opt-in decision-capture nudge (nudge-class).

decision_reconcile_nudge

Stop-event NUDGE for Decision Register drift.

descriptive_name

advise against generic file names (nudge class).

dev_override_wiring_nudge

warn when a `.harness-dev/<name>.yaml` override is present but wired to no `HARNESS_*` env, so it is silently ignored (the loader falls back to the shipped default).

discover_isolation_nudge

advisory /clear-before-plan reminder (nudge class).

engine_root_inject

SessionStart engine-root context (telemetry, fail-open).

engine_skew_nudge

SessionStart courier health advisory (nudge, fail-open).

gemini_stop_review_gate

Stop-event hook: an optional gemini review of the assistant's last turn. REWRITTEN fail-open (NOT the vendored fail-closed .mjs).

glossary_capture_nudge

opt-in glossary-capture nudge (nudge-class).

goal_cycle_nudge

advisory reminder to drop a cycle-memory breadcrumb at a tick boundary (nudge class).

instruction_policy_nudge

PreToolUse nudge: say what the project's adopted instruction policy says about this edit or this landing.

ledger_red_watch

PostToolUse:{Write|Edit|MultiEdit, Bash} nudge (nudge-class).

memory_gap_hook

opt-in tier-1 memory-gap hook (nudge-class).

memory_index_watch

PostToolUse:{Write|Edit|MultiEdit, Bash} nudge (nudge-class).

mutation_guard

SubagentStart/Stop DETECTION hook (nudge class, fail-open).

pending_decisions_resurface

SessionStart hook: after a compaction, re-surface the most-recent still-open AskUserQuestion so a summary cannot quietly bury a decision the user was still weighing.

plan_dir_nudge

PreToolUse(Write/Edit/MultiEdit) nudge: a write under a `plans/<dir>/` that has no valid `plan.md` will never anchor a hard stage.

rule_nudge_hook

name the rules whose scope matches the file being written.

spawn_fingerprint_audit

Ask-3 C3's group-fingerprint post-run audit (nudge-class, OBSERVATION-only).

stale_plan_close_nudge

advisory reminder to close a finished-but-open plan before it gates unrelated work (nudge class).

standards_drift_nudge

keep the auto-loaded prose standards from drifting.

usage_nudge

advisory near-ceiling reminder before a heavy spawn (nudge class).

yaml_comment_cut_nudge

PostToolUse:{Write|Edit|MultiEdit} nudge (nudge-class).

zsh_split_nudge

PreToolUse(Bash) nudge: an unquoted multi-word parameter expansion is NOT word-split under zsh.

telemetry

TelemetryĐo lường (21)

fail-open, silent — records what ran; an internal error never blocks a tool.

agent_memory_stray

SubagentStop telemetry hook: remove the empty `.claude/agent-memory/<agent>/` dirs Claude Code leaves below the project root.

auto_finalize_ship

PreToolUse(Skill) hook: ship-belt backstop that finalizes the single active plan when N/N phases have a PASS snapshot.

bash_write_guard

PostToolUse:Bash advisory (telemetry-class, fail-open).

emit_session_summary

on session Stop, emits one line to state/telemetry/sessions.jsonl (telemetry-class):

glossary_pointer_inject

SessionStart hook (telemetry-class) injecting a one-line pointer at the glossary as additionalContext.

inject_prompt_context

context re-injection hook (telemetry-class).

manual_test_anchor

PostToolUse:Bash telemetry hook (manual-test anchor).

mark_bash_start

PreToolUse:Bash hook (telemetry-class).

notify_remote

Notification-event hook: relay Claude's notification to a user-configured webhook so an UNATTENDED run (/goal until-429) reaches the operator's phone/chat the moment something needs them.

nudge_context_inject

UserPromptSubmit hook (telemetry-class) re-surfacing turn-end nudge observations to the MODEL as additionalContext.

panel_flag_relay

mechanical `--ultra`/`--debate` panel-flag detector (UserPromptSubmit, telemetry-class, `kind: additionalContext`).

phase_progress_writer

PostToolUse hook: per-phase evidence snapshot + plan lifecycle (auto-open, auto-close).

reinject_stop_context

Stop-event hook: re-inject working context on a /goal continuation.

session_init

SessionStart hook (telemetry-class).

setup_nudge

SessionStart onboarding advisory (telemetry-class, fail-open).

spawn_ledger_writer

append each Agent/Task spawn to the provenance ledger (telemetry, fail-open).

stamp_artifact

PostToolUse hook: stamp harness provenance into plan/ report markdown as it is written (telemetry-class).

subagent_init

SubagentStart context injector (telemetry, fail-open).

track_script_execution

PostToolUse:Bash hook (telemetry-class).

track_skill_invocation

records every skill invocation to state/telemetry/invocations.jsonl (telemetry-class).

track_subagent_outcome

SubagentStop hook (telemetry-class).