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.
npx agentmods add commands/open-latch/latch/unlatchgit clone --depth 1 https://github.com/open-latch/latchWhat 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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00008 | $0.00617 |
| Opus 5 | $0.00004 | $0.00309 |
| Sonnet 5 | $0.00002 | $0.00123 |
| Haiku 4.5 | $0.00001 | $0.00062 |
Grade A, and why
unlatch 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 3d 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.
How it starts
The opening of the file, as written. The whole thing — 75 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Use this when the user wants latch out of the way, wants to compare against vanilla agent behavior, or wants to re-latch after using the escape hatch.
Unlatched mode is in-place and install-level. It does not clone the project,
delete KB data, uninstall latch, call a cloud service, or collect telemetry. It
creates a visible UNLATCHED receipt plus the normal full-disable sentinel so
latch's automatic project-judgment layer is off for this latch install until the
user re-latches. If the user changes repos before re-latching, latch remains off
and should say so loudly. It also masks latch's managed CLAUDE.md /
AGENTS.md regions in the current project/ancestor files while unlatched, then
restores them when re-latched, so native project-instruction loading does not
keep carrying latch's contract in the repo where the user unlatched.
Never mutate state immediately from this command, even in dangerous/yolo permission modes. First inspect state and ask for explicit confirmation.
Inspect
bash <KB_HOME>/bin/unlatch.sh
If the output says LATCHED, tell the user:
Latch is currently LATCHED.
Switch to UNLATCHED mode?
This turns latch's project-judgment layer off for this latch install, masks latch-managed CLAUDE.md/AGENTS.md regions in this project, and leaves KB data intact.
Latch remains off for this latch install, even if you change repos, until you re-latch.
To re-latch later, run /unlatch again.
Reply exactly: unlatch
If the output says UNLATCHED, tell the user:
Latch is currently UNLATCHED.
Switch back to LATCHED mode?
Latch hooks will resume on the next prompt unless LATCH_UNLATCHED is set.
If LATCH_UNLATCHED is set, unset it too.
Reply exactly: latch
Stop there. Do not run the state-changing command until the user replies with the exact confirmation word.
Confirmed Action
Only after the user replies exactly unlatch:
bash <KB_HOME>/bin/unlatch.sh --confirm unlatch
Only after the user replies exactly latch:
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.
- 3d ago First seen · 75 lines · 8 tokens per session scan A c1b75237f908
unlatch is a command published in the GitHub repository open-latch/latch (9 stars, last pushed 3d ago), licensed Apache-2.0. It adds 8 tokens to every session and 617 once invoked, about $0.0000 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.
Other commands, from other repositories
ctx
Search agent history or trace code to its original agent session.
add-agent
引导新增一个 Agent 适配器。用法 /add-agent.
check
跑全套本地校验(前端 lint/test/build + Rust check)。.
handoff
Export a Waybill Bundle by default, or import one when requested.
handoff-import
Read .claude/skills/handoff/SKILL.md completely and dispatch its import workflow. Use any supplied command arguments as the bundle path, defaulting to .waybill/; preserve the skill's read-only and untrusted-data boundaries.
make-this
Install the Multi-Agent Loop Kit into the current repo and run the setup interview.