Guide · Agent skill
How to use the freeze skill to lock a decision for a cooldown
Updated · written with AI from the project's own documentation and checked against our catalogue data
The freeze skill is an agent skill for Claude-compatible agents that locks a strategic decision for a set cooldown period so it is not reopened on impulse. It is aimed at founders and executives who have made a hard-to-reverse decision under pressure and want a written cooling-off period before acting on it. After this guide you will know what the skill does, how to install it, how to ask for a freeze in plain language, and what it cannot guarantee.
Decision Cooldown Lockalirezarezvani/claude-skills · MIT Download the skill Quick start
- 1Confirm the requirements. Check that you have a Claude-compatible agent with the c-level-agents plugin installed, and write access to ~/.claude/freezes/.
- 2Download and install the ZIP. Download the ZIP from the page and unzip it into ~/.claude/skills/ or a project's .claude/skills/, or upload it under Settings, Capabilities, Skills.
- 3Record the decision. Make sure the decision has a record with APPROVED status, since the workflow checks for it before applying a freeze.
- 4Trigger the freeze. Run /c-level-agents:freeze with the decision path and number of days, or leave the days out to use the default for that decision type.
- 5Check the record and index. Confirm the decision record shows FROZEN with the frozen-until date and that ~/.claude/freezes/active.md lists it.
- 6Release only with a reason. If you must reverse early, run the release command with a written reason, knowing the override is logged.
What does the freeze skill make the agent do?
The skill marks a recorded decision as frozen until a date you set. While it is frozen, the chief-of-staff router is meant to refuse to send the same topic back to the boardroom for re-debate, unless a pre-set kill criterion triggers. The skill is adapted from gstack's /freeze and /guard safety primitives, changed from code scoping to strategic scoping.
Its stated purpose is to stop decision fatigue from producing reversals. The freeze is described as a process, not a rule: every override is logged so a later post-mortem can review it.
- Writes FROZEN status and a freeze_until date into the decision record
- Adds the decision to an active-freezes index file
- Blocks re-routing to the boardroom until expiry or a kill criterion fires
- Logs any early release with a stated reason
How do you install the freeze skill?
Download the ZIP for this skill from our page. Unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder to share it with one project. Alternatively, upload the ZIP in the Claude apps under Settings, then Capabilities, then Skills.
The package contains a single file, SKILL.md. The skill refers to other components, including the cs-chief-of-staff agent, the decision-logger skill and the post-mortem command, which are not included in this file. You need the c-level-agents framework installed for those references to work.
- Personal install: ~/.claude/skills/
- Project install: .claude/skills/ inside the project folder
- Apps install: Settings, then Capabilities, then Skills, then upload the ZIP
How do you trigger the freeze with a natural request?
The documented command is /c-level-agents:freeze followed by a decision reference and a number of days. The skill's description lists the arguments as [decision] [days], while the body lists them as <decision-path> <days>, so use the path to the decision record you actually have. If you do not specify a period, the defaults below apply.
The skill description says it is for irreversible decisions made under pressure, such as a layoff plan or a multi-year contract, before execution. You can also ask in plain language that names the decision and the cooling-off intent, and the agent should map that to the same workflow.
- Default freeze: fundraise round size or lead choice, 30 days
- Default freeze: pricing change, 60 days
- Default freeze: market entry or exit, 90 days
- Default freeze: layoff or RIF, 30 days
- Default freeze: strategic pivot, 90 days
- Default freeze: personnel exec hire or fire, 60 days
- Default freeze: M&A LOI, 30 days
- Custom: specify the period in the command
/c-level-agents:freeze <decision-path> <days>What happens during a freeze, and how do you release one?
The workflow reads the decision record, checks that it has APPROVED status, writes freeze_until into the record, and adds it to ~/.claude/freezes/active.md. The record then shows the frozen status, the reason and the override condition.
To end a freeze early, the founder runs the release command with a reason. The release is written to the decision history and stays there. If a kill criterion in the decision fires, the freeze releases automatically and the router sends the matter to the post-mortem command.
- Expiry: the freeze ends on the frozen-until date
- Early release: founder runs release with a stated reason
- Auto-release: a kill criterion triggers and routes to post-mortem
- Blocked routing: the boardroom command is unavailable until release or expiry
/c-level-agents:freeze release <decision> <reason>What are the safety notes and limits?
The skill runs no scripts and makes no network calls, according to its listing. It needs write access to ~/.claude/freezes/ to maintain the active-freezes index.
The freeze is enforced by the agent following the instructions in the file, not by a technical lock. It can be bypassed or ignored, so treat it as a written commitment and a discipline aid rather than a guarantee. Because it is designed to resist reversal, it can also delay urgent operational or safety changes, so do not use it for those.
- No scripts run and no network use
- Needs write access to ~/.claude/freezes/
- Not a system-level lock
- Unsuitable for urgent operational or safety issues
What people use it for
- Holding a layoff plan for 30 days after it was agreed under budget pressure before any notices go out
- Locking a multi-year vendor contract decision so the team executes it rather than reopening it each week
- Preserving a split-vote boardroom outcome from repeated second-guessing during a personnel transition
- Protecting a pricing change for 60 days so results can be judged before another debate
Frequently asked questions
Is the freeze skill free to use and what licence does it have?
The listing states the licence is MIT. The source does not state any price, so check the directory page for any terms beyond the licence.
What do I need to run the freeze skill?
You need a Claude-compatible agent with the c-level-agents plugin installed, because the chief-of-staff agent, decision-logger skill and post-mortem command are referenced but not included. You also need write access to ~/.claude/freezes/.
Where do I install the freeze SKILL.md file?
Unzip the downloaded ZIP into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder. You can also upload the ZIP in the Claude apps under Settings, Capabilities, Skills.
Does the freeze skill actually stop a decision from being changed?
Not technically. The agent is instructed to refuse to reopen the debate, but the freeze relies on the agent following the file, so it can be bypassed or ignored. Overrides are logged for review.
Can I end a freeze early?
Yes. The founder runs the release command with a stated reason, and that release is recorded in the decision history. A kill criterion that triggers also releases the freeze automatically.
Can the freeze skill be used for urgent problems?
The source says no. The lock is designed to resist reversal and can delay needed changes, so urgent operational or safety issues should not be frozen.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
The freeze skill suits founders and executives who want a written cooling-off period after a major decision made under pressure, especially within the c-level-agents setup. It is a discipline aid that depends on the agent following instructions, so people who need a hard technical guarantee, or who face urgent decisions, should choose another approach.
Download the skill