latch-gate

A manually requested go/no-go review for a coding or build request. It returns one of four recommendations: proceed, modify, do not proceed, or ask a person to decide.

In plain words
What is it for?
Use it to review a proposed implementation, see the stored decisions supporting the recommendation, and get a suggested next action.
Why use it?
It provides an explicit second opinion before work begins, while leaving the final choice and any redirection to the user or agent.

Command

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add commands/open-latch/latch/latch-gate
Clone the repo
git clone --depth 1 https://github.com/open-latch/latch
Per session 15 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 449 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00015 $0.00449
Opus 5 $0.00008 $0.00225
Sonnet 5 $0.00003 $0.00090
Haiku 4.5 $0.00002 $0.00045

Measured 2d ago against content hash 412931765ac1, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

latch-gate scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

commands/latch-gate.md · 38 lines

What it actually says

Manually invoke the per-project gate (heavyweight LLM go/no-go judgement layer) for the request in $ARGUMENTS. The primary MCP tool is latch_gate; kb_gate remains a legacy alias. This slash is the manual escape hatch when you want an explicit second opinion without relying on auto-invocation.

The classifier returns one of {PROCEED | MODIFY | DO_NOT_PROCEED | NEEDS_HUMAN_JUDGMENT} with cited node ids, a recommended better-next-action, and a side-note summary. It does NOT auto-redirect — the user/agent decides.

Legacy aliases: /kb-gate and kb_gate may exist on older installs and remain supported for compatibility. Prefer /latch-gate and latch_gate in fresh instructions and docs.

Budget-gated (counts against the daily LLM cap shared with /latch-compact and nightly heal). Skip and report the budget message if the cap is hit.

Run

bash <KB_HOME>/bin/run_latch_gate.sh "$ARGUMENTS"

After the command returns, show an explicit Latch gate block. Prefer the returned findings object; it is already shaped for chat display and should be shown even when the recommendation is PROCEED or the gate was skipped. If the result says latch is UNLATCHED, report that latch gate was skipped, do not claim KB evidence was read, and tell the user to run /unlatch to re-latch. Otherwise, lead with provenance that Latch ran the gate on the request/plan, then show the verdict, summary/rationale, receipt / source basis when present, cited evidence nodes with status/current authority, better_next_action, and uncovered claims. Don't re-render the full chains object — point the user at the cited ids and offer to latch_get any of them for full bodies.

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 38 lines · 15 tokens per session scan A 412931765ac1

Subscribe to this mod's changes

latch-gate is a command published in the GitHub repository open-latch/latch (9 stars, last pushed 3d ago), licensed Apache-2.0. It adds 15 tokens to every session and 449 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.