Update to Atbash Safety
Snapshot Nov 22, 2026 · 12:12 UTC · version 0.3.5
Collection source: downloaded plugin package.
Payment or plan references changed
Full description — Now says: troubleshoot, or switch the Atbash Safety profile for Codex. Use when the user wants to sign up for Atbash, connect an existing Atbash agent, create a new...
Observed in published text. Live prices and checkout terms have not been verified by this change.
Up to 3 plugins free · no card required.
All fields and original values.
Full description
Nov 2, 2026 · 12:47 UTC
or switch the local Atbash agent used for explicit safety checks. Use for Atbash signup, existing-agent connection, setup progress, or local profile troubleshooting.
Nov 22, 2026 · 12:12 UTC
troubleshoot, or switch the Atbash Safety profile for Codex. Use when the user wants to sign up for Atbash, connect an existing Atbash agent, create a new agent, check setup progress, or repair local plugin configuration.
Skill instructions
Nov 2, 2026 · 12:47 UTC
or switch the local Atbash agent used for explicit safety checks. Use for Atbash signup, existing-agent connection, setup progress, or local profile troubleshooting. Run the bundled helper yourself in a local desktop thread. It requires ...
Nov 22, 2026 · 12:12 UTC
troubleshoot, or switch the Atbash Safety profile for Codex. Use when the user wants to sign up for Atbash, connect an existing Atbash agent, create a new agent, check setup progress, or repair local plugin configuration. Use the bundled...
Supporting files
Nov 2, 2026 · 12:47 UTC
[{"relative_path":"agents/openai.yaml","size_in_bytes":207},{"relative_path":"scripts/atbash-control.mjs","size_in_bytes":551}]
Nov 22, 2026 · 12:12 UTC
[{"relative_path":"agents/openai.yaml","size_in_bytes":183},{"relative_path":"scripts/atbash-control.mjs","size_in_bytes":551},{"relative_path":"scripts/prepare-config.mjs","size_in_bytes":2788}]
Compare saved observations
Download comparison JSONSKILL.md text changes
− removed · + added
--- before +++ after @@ -1,46 +1,121 @@ --- name: atbash-setup -description: Set up, connect, verify, or switch the local Atbash agent used for explicit safety checks. Use for Atbash signup, existing-agent connection, setup progress, or local profile troubleshooting. +description: Set up, connect, verify, troubleshoot, or switch the Atbash Safety profile for Codex. Use when the user wants to sign up for Atbash, connect an existing Atbash agent, create a new agent, check setup progress, or repair local plugin configuration. --- # Atbash Setup -Run the bundled helper yourself in a local desktop thread. It requires Node.js 22.13.0 or newer on macOS arm64, Linux x64/arm64 with glibc, or Windows x64. Resolve `<skill-directory>` to the directory containing this file; use its installed launcher throughout a job. If local execution is unavailable, explain that this package requires a local desktop thread. +Use the bundled control helper for onboarding. Keep private keys outside the conversation and outside tool arguments. -## Onboard an agent +## Carry out setup + +Treat a request to sign up, set up, or connect Atbash as a request to perform the workflow. Run the bundled commands yourself with the available execution tool; command examples below are for you to execute, not instructions to hand back to the user. Resolve `<skill-directory>` to the directory containing this SKILL.md, and use that installed copy's launcher throughout the job. + +Handle session creation, progress checks, discovery, plan-file creation and submission, and local activation yourself. Ask only for missing user choices and actions requiring the user's identity or consent: wallet sign-in, review and signature of the exact proposal, local entry of an existing agent key, and enabling or trusting the host hook. Do not send the user to manually create an account, organization, or agent when the control API supports that step. + +After a user completes verification or approval, inspect the same job and perform the next available step. Keep the job ID in the conversation so setup can resume without asking the user to run commands. Do not claim completion until the helper and status checks confirm it. + +## Start or resume setup + +The helper runs outside the Codex sandbox. It must resolve and reach the Atbash dashboard, the chain nodes and its own loopback key page, and write the Atbash configuration directory (`~/.config/atbash`, or `$ATBASH_CONFIG_DIR` when it is set). Inside the sandbox, even with network access granted, its DNS lookups fail, so do not try it sandboxed first and do not use `request_permissions` or `additional_permissions` for it. + +Run every helper command, from the first one, with these shell tool parameters: + +- `sandbox_permissions`: `"require_escalated"` +- `justification`: one sentence saying the Atbash setup helper needs to reach Atbash and save its configuration outside the sandbox. +- `prefix_rule`: `["node", "<skill-directory>/scripts/atbash-control.mjs"]`, the same absolute launcher path as the command, so the user can approve the helper once instead of on every step. + +If the shell tool has no `sandbox_permissions` parameter, Codex is already running without a sandbox: run the command as usual. If Codex rejects the escalation because its approval policy never asks, tell the user to switch Codex's permissions to Full access with `/permissions`, then run the same command again. A helper error that says Atbash "could not be reached" or "could not write" means the command ran inside the sandbox, not that the service is down: run the same command again once with the parameters above. + +Run the helper through this skill's `scripts/atbash-control.mjs` launcher, using the absolute skill directory: ```text -node <skill-directory>/scripts/atbash-control.mjs setup start --host codex +node "<skill-directory>/scripts/atbash-control.mjs" setup start --host codex ``` -The helper uses the `codex` host profile namespace in both ChatGPT desktop and Codex. Show the returned `verificationUri` and code. The user signs in and verifies their wallet in Connect Atbash. Keep the job ID to resume the same job. Ask only for missing choices and browser approval; perform helper commands and plan-file creation yourself. +Until setup activates a profile, the hook allows only these setup steps, run outside the sandbox as above, and questions to the user, and denies everything else with "Atbash is not set up yet". Run each helper command as one plain command in exactly this form: no `cd`, `&&`, `;`, pipes, redirection, environment-variable prefixes, or command substitution, and no `--service` option. Commands in any other form are denied. -Run `setup inspect <job-id>` through the same launcher after wallet verification. Follow `nextAction`: +The result contains a public `verificationUri`, verification code, opaque job ID, and `planPath`. Show the URL and code to the user and ask them to complete wallet verification in **Connect Atbash**, then come back to the conversation. Never expose files under `~/.config/atbash/pending`, `credentials`, `profiles`, or `hosts`. -- `OPEN_BROWSER`: wait for the user to verify their wallet at the returned URL. -- `PREPARE_PLAN`: use discovery and user-provided choices to create a non-secret JSON file with an `actions` array. Submit it with `setup plan <job-id> --input <path>`. -- `REVIEW_IN_BROWSER`: show the verification URL; the user reviews and signs the exact proposal. Do not approve or sign for them. -- `WAIT`: inspect again after two seconds, keeping the user informed. Stop on expiry or terminal failure; do not repeatedly create new sessions. -- `ACTIVATE`: run `setup continue <job-id>` to decrypt the key locally and select the agent profile. -- `RECOVER`: report completed and failed steps. Further mutations require a new authorization session. -- `DONE`: run `node <skill-directory>/../../runtime/public-safety.cjs status`. Report whether the agent is ready for explicit safety checks. +Inspect progress with `setup inspect <job-id>` through the same launcher. Follow `nextAction`: -A new-agent plan uses `create_account` and `create_organization` when missing, `activate_free_plan` if needed, then one `create_agent` with `keySource: "generate_in_browser"`. Include `network`, `organization`, `name`, `purpose`, and `risk` (`low`, `medium`, `high`, or `critical`) from discovery and the user's choices. Account actions have only `type` and `network`; organization creation uses `name` and `description`. Free-plan activation uses `organization` and requires the public network. +- `OPEN_BROWSER`: the user completes sign-in and wallet verification in the provided page. +- `PREPARE_PLAN`: use discovery to determine what exists. Ask for any missing names or choices, then build a non-secret plan JSON containing only `actions` yourself and pass it to the launcher inline, as one single-quoted argument: `setup plan <job-id> --json '{"actions":[...]}'`. Do not write it to a file first — writing a file means running a shell command, which setup mode does not allow. Do not ask the user to write JSON or run the command. +- If the organization has no active subscription, run `setup checkout <job-id> --organization <name>`. Show the returned `checkoutUrl` and ask the user to choose a plan in the dashboard's existing checkout flow. Resume inspection after the dashboard returns to `returnUrl`; never select paid plans or handle payment details yourself. +- `REVIEW_IN_BROWSER`: the user reviews and signs the exact proposal in Connect Atbash. Do not approve it for them. +- `WAIT`: inspect again after the returned poll interval, or after two seconds if none is present. Keep the user informed during longer waits; stop on expiry or a terminal failure. +- `ACTIVATE`: run `setup continue <job-id>` to decrypt the delivered key locally and activate the profile. +- `RECOVER`: report completed and failed steps. Any replacement mutation requires a new management session and approval. +- `DONE`: `setup continue` already reports the new agent's `agentStatus`. Report setup as finished only when `nextAction` is `DONE` and `agentStatus.state` is `ready`. A session `status` of `completed` alone does not mean the profile is active. Do not run another command to check status: from this point every tool call is judged under the new agent's policy, and a blocked call can jail the agent. To confirm enforcement, make one harmless call that fits the agent's stated purpose. + +Registering the hook is the user's step. If their `/hooks` list does not show the Atbash `PreToolUse` hook, ask them to run `node <skill-directory>/../../runtime/install-hook.cjs` from their own terminal, restart Codex, and trust the hook. Do not run the installer yourself: where the plugin's own hook already runs it registers a second copy, and once setup activates a profile it is judged like any other action. The plugin's bundled `hooks/hooks.json` entry runs a bare `node` and a `PLUGIN_ROOT` placeholder: it depends on `node` being on the host's PATH and on `PLUGIN_ROOT` being exported by the host, so it is not a gate on its own. + +For a new setup, the plan normally contains `create_account` when missing, `create_organization` when missing, `activate_free_plan` when no subscription exists, then `create_agent` with `keySource: "generate_in_browser"`. Use only values the user supplied or explicitly chose. Do not invent organization names, organization descriptions, purposes, risks, or agent names; ask for any that are missing before building the plan. + +### Plan format + +A plan is `{"actions": [...]}` with the actions in the order below. There are no other top-level fields. Every action carries its own `"network"` (`"public"` or `"private"`), and all actions in one plan use the same network. Use a network listed in discovery's `availableNetworks`, which names the networks this Atbash deployment serves: when it lists only one, use it without asking; when it is absent, ask the user. Use exactly these fields: + +| Action | Include when | Fields | +| --------------------- | ------------------------------------------ | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | +| `create_account` | discovery shows no account on that network | `type`, `network` | +| `create_organization` | the organization does not exist yet | `type`, `network`, `name` (2–64 letters, digits, `.`, `_`, or `-`), `description` (1–280 characters) | +| `activate_free_plan` | the organization has no subscription | `type`, `network`, `organization` (the organization's name) | +| `create_agent` | always, for a new agent | `type`, `network`, `organization`, `name` (up to 80 characters), `purpose` (up to 280), `risk` (`low`, `medium`, `high`, or `critical`), `keySource` (`"generate_in_browser"`), `policy` | + +For example, a new user on a deployment that serves only the private network: + +```json +{ + "actions": [ + { "type": "create_account", "network": "private" }, + { + "type": "create_organization", + "network": "private", + "name": "acme_ai", + "description": "Acme agents" + }, + { "type": "activate_free_plan", "network": "private", "organization": "acme_ai" }, + { + "type": "create_agent", + "network": "private", + "organization": "acme_ai", + "name": "codex dev", + "purpose": "Coding assistant for the Acme repository", + "risk": "medium", + "keySource": "generate_in_browser", + "policy": { "preset": "developer", "version": 1 } + } + ] +} +``` -Suggest `policy: {"preset":"developer","version":1}` for coding, `devops` for operations, or `full_baseline` when the user requests the strictest protection. Explain the suggestion before approval; the user can choose a different preset or customize the policy in Connect Atbash. +Activate the Free plan on the same network as the rest of the plan. If the dashboard answers that the Free plan is available only on the public network, that deployment keeps Free on public: a private organization there needs a paid plan first, so offer `setup checkout` for an existing organization or suggest the public network. Because the plan is passed in single quotes, write any single quote inside a value as `\u0027` (for example `"Nati\u0027s agent"`). If the helper rejects the plan, its error names the action number and the field to fix; correct that field and submit the whole plan again. -If an existing organization needs a subscription, run `setup checkout <job-id> --organization <name>` and show the returned checkout URL. The user selects the plan and handles payment in their browser. Resume inspection afterward. +Give every new agent a starting policy by adding `"policy": { "preset": "<preset>", "version": 1 }` to its `create_agent` action. Choose the preset from the agent's purpose: `developer` for coding work (the usual choice for Codex), `devops` for infrastructure, deployment, or operations work, and `full_baseline` when the user asks for the strictest protection. Tell the user which preset you suggested and that they can switch it or customize individual rules on the Connect Atbash review page before approving; the policy they approve is created as the organization policy "<agent name> Policy" and can be fine-tuned later under Policies in the dashboard. If the user wants a different preset before review, resubmit the plan with it. ## Connect an existing agent -After wallet verification and discovery, run `profile connect <job-id>` through the launcher. Show the returned loopback `localUri`. The user enters their key only in that local form; the helper verifies that its public identity belongs to the verified wallet. Wait for the helper's completion, then run the status command above. +After wallet verification and discovery, run `profile connect <job-id>` through the launcher. Show the returned loopback `localUri` to the user. If the user abandons the connection, or the local page expires before a key is entered, cancel the job as described under **Cancel setup**. The user enters the key in that local page. The helper derives its public key and connects it only if discovery shows the verified wallet owns the matching agent. The private key is never sent to Atbash or printed. + +Never ask the user to paste, upload, reveal, or dictate a private key. Never read or print credential files. Never put a key in a plan, prompt, environment assignment, command-line argument, committed file, or shell history. + +## Cancel setup + +`setup cancel <job-id>`, run through the same launcher, ends a setup job: it cancels the session in Connect Atbash, so its link stops working, and deletes the job's local files, including its session secret. Setup mode allows it, and like the other helper commands it runs outside the sandbox. + +- Run it when the user asks to stop, cancel, or start over, or abandons setup, while `nextAction` is `OPEN_BROWSER`, `PREPARE_PLAN`, or `REVIEW_IN_BROWSER`. Tell the user setup was cancelled. To start over, run `setup start` for a new job. +- Run it at `RECOVER` when the session was cancelled, rejected, or expired, and for an expired job you will not resume, so its secrets do not stay on disk. If the helper reports that the session had already ended on Atbash, the cleanup still succeeded. +- Do not cancel after the user approved the proposal: approved steps run in the browser and cancelling does not undo them. Let the job finish and report what was created. At `ACTIVATE` the new agent's key is waiting to be activated and the helper refuses to cancel; run `setup continue <job-id>` instead. + +## Profiles and legacy compatibility -Never ask for, read, print, or put an agent private key in the conversation, tool arguments, environment assignments, or plan files. Credential and pending-session files are private. A browser verification URL and job ID may be shown; session secrets may not. +List profiles with `profile list --host codex`, select one with `profile switch --host codex --profile <id>`, or disconnect the host mapping with `profile disconnect --host codex`. Disconnecting retains the credential; it does not delete or revoke the on-chain agent. -## Profiles and troubleshooting +If there is no selected profile, the hook keeps the legacy SDK configuration behavior. If `ATBASH_AGENT_KEY` or `ATBASH_ORG_NAME` conflicts with a selected profile, report the conflict and ask the user to remove or correct the override locally. Never inspect the conflicting key. -Use the same launcher for `profile list --host codex`, `profile switch --host codex --profile <id>`, and `profile disconnect --host codex`. Disconnecting retains the stored credential and does not revoke the on-chain agent. After switching, run status again. Resolve an environment mismatch by connecting an agent belonging to the service reported by `public-safety.cjs help`; do not copy keys between environments. +Setup runs with the hook trusted; do not ask the user to disable the plugin, which also removes this skill. While no configuration exists at all, the hook allows the setup steps above and denies everything else. -Report helper failures without exposing credential contents or requesting backend credentials. If a local environment override conflicts with a profile, ask the user to correct it locally without revealing its value. +If the hook denies a setup step because a configuration already exists (invalid, jailed, or not registered), setup mode does not apply — report the exact denial and tell the user they can disconnect the current profile from their own terminal with `node "<skill-directory>/scripts/atbash-control.mjs" profile disconnect --host codex`, then start setup again. Do not describe this as bypassing an individual verdict. -This package checks only explicitly submitted actions. Agent readiness does not mean other tool calls are automatically intercepted. After setup, use `$atbash-safety` when the user requests an action check. +If the API fails, report the failing step and the returned error. A backend error is not a reason to hand the same command to the user or ask them to supply backend credentials. Do not repeatedly create sessions or claim that a manual command will fix a server failure.
Full technical diff · 3 changed fields
changed /description
"Set up, connect, verify, or switch the local Atbash agent used for explicit safety checks. Use for Atbash signup, existing-agent connection, setup progress, or local profile troubleshooting."
"Set up, connect, verify, troubleshoot, or switch the Atbash Safety profile for Codex. Use when the user wants to sign up for Atbash, connect an existing Atbash agent, create a new agent, check setup progress, or repair local plugin configuration."
changed /included_files
[
{
"relative_path": "agents/openai.yaml",
"size_in_bytes": 207
},
{
"relative_path": "scripts/atbash-control.mjs",
"size_in_bytes": 551
}
][
{
"relative_path": "agents/openai.yaml",
"size_in_bytes": 183
},
{
"relative_path": "scripts/atbash-control.mjs",
"size_in_bytes": 551
},
{
"relative_path": "scripts/prepare-config.mjs",
"size_in_bytes": 2788
}
]changed /skill_md_contents
"---\nname: atbash-setup\ndescription: Set up, connect, verify, or switch the local Atbash agent used for explicit safety checks. Use for Atbash signup, existing-agent connection, setup progress, or local profile troubleshooting.\n---\n\n# Atbash Setup\n\nRun the bundled helper yourself in a local desktop thread. It requires Node.js 22.13.0 or newer on macOS arm64, Linux x64/arm64 with glibc, or Windows x64. Resolve `<skill-directory>` to the directory containing this file; use its installed launcher throughout a job. If local execution is unavailable, explain that this package requires a local desktop thread.\n\n## Onboard an agent\n\n```text\nnode <skill-directory>/scripts/atbash-control.mjs setup start --host codex\n```\n\nThe helper uses the `codex` host profile namespace in both ChatGPT desktop and Codex. Show the returned `verificationUri` and code. The user signs in and verifies their wallet in Connect Atbash. Keep the job ID to resume the same job. Ask only for missing choices and browser approval; perform helper commands and plan-file creation yourself.\n\nRun `setup inspect <job-id>` through the same launcher after wallet verification. Follow `nextAction`:\n\n- `OPEN_BROWSER`: wait for the user to verify their wallet at the returned URL.\n- `PREPARE_PLAN`: use discovery and user-provided choices to create a non-secret JSON file with an `actions` array. Submit it with `setup plan <job-id> --input <path>`.\n- `REVIEW_IN_BROWSER`: show the verification URL; the user reviews and signs the exact proposal. Do not approve or sign for them.\n- `WAIT`: inspect again after two seconds, keeping the user informed. Stop on expiry or terminal failure; do not repeatedly create new sessions.\n- `ACTIVATE`: run `setup continue <job-id>` to decrypt the key locally and select the agent profile.\n- `RECOVER`: report completed and failed steps. Further mutations require a new authorization session.\n- `DONE`: run `node <skill-directory>/../../runtime/public-safety.cjs status`. Report whether the agent is ready for explicit safety checks.\n\nA new-agent plan uses `create_account` and `create_organization` when missing, `activate_free_plan` if needed, then one `create_agent` with `keySource: \"generate_in_browser\"`. Include `network`, `organization`, `name`, `purpose`, and `risk` (`low`, `medium`, `high`, or `critical`) from discovery and the user's choices. Account actions have only `type` and `network`; organization creation uses `name` and `description`. Free-plan activation uses `organization` and requires the public network.\n\nSuggest `policy: {\"preset\":\"developer\",\"version\":1}` for coding, `devops` for operations, or `full_baseline` when the user requests the strictest protection. Explain the suggestion before approval; the user can choose a different preset or customize the policy in Connect Atbash.\n\nIf an existing organization needs a subscription, run `setup checkout <job-id> --organization <name>` and show the returned checkout URL. The user selects the plan and handles payment in their browser. Resume inspection afterward.\n\n## Connect an existing agent\n\nAfter wallet verification and discovery, run `profile connect <job-id>` through the launcher. Show the returned loopback `localUri`. The user enters their key only in that local form; the helper verifies that its public identity belongs to the verified wallet. Wait for the helper's completion, then run the status command above.\n\nNever ask for, read, print, or put an agent private key in the conversation, tool arguments, environment assignments, or plan files. Credential and pending-session files are private. A browser verification URL and job ID may be shown; session secrets may not.\n\n## Profiles and troubleshooting\n\nUse the same launcher for `profile list --host codex`, `profile switch --host codex --profile <id>`, and `profile disconnect --host codex`. Disconnecting retains the stored credential and does not revoke the on-chain agent. After switching, run status again. Resolve an environment mismatch by connecting an agent belonging to the service reported by `public-safety.cjs help`; do not copy keys between environments.\n\nReport helper failures without exposing credential contents or requesting backend credentials. If a local environment override conflicts with a profile, ask the user to correct it locally without revealing its value.\n\nThis package checks only explicitly submitted actions. Agent readiness does not mean other tool calls are automatically intercepted. After setup, use `$atbash-safety` when the user requests an action check.\n""---\nname: atbash-setup\ndescription: Set up, connect, verify, troubleshoot, or switch the Atbash Safety profile for Codex. Use when the user wants to sign up for Atbash, connect an existing Atbash agent, create a new agent, check setup progress, or repair local plugin configuration.\n---\n\n# Atbash Setup\n\nUse the bundled control helper for onboarding. Keep private keys outside the conversation and outside tool arguments.\n\n## Carry out setup\n\nTreat a request to sign up, set up, or connect Atbash as a request to perform the workflow. Run the bundled commands yourself with the available execution tool; command examples below are for you to execute, not instructions to hand back to the user. Resolve `<skill-directory>` to the directory containing this SKILL.md, and use that installed copy's launcher throughout the job.\n\nHandle session creation, progress checks, discovery, plan-file creation and submission, and local activation yourself. Ask only for missing user choices and actions requiring the user's identity or consent: wallet sign-in, review and signature of the exact proposal, local entry of an existing agent key, and enabling or trusting the host hook. Do not send the user to manually create an account, organization, or agent when the control API supports that step.\n\nAfter a user completes verification or approval, inspect the same job and perform the next available step. Keep the job ID in the conversation so setup can resume without asking the user to run commands. Do not claim completion until the helper and status checks confirm it.\n\n## Start or resume setup\n\nThe helper runs outside the Codex sandbox. It must resolve and reach the Atbash dashboard, the chain nodes and its own loopback key page, and write the Atbash configuration directory (`~/.config/atbash`, or `$ATBASH_CONFIG_DIR` when it is set). Inside the sandbox, even with network access granted, its DNS lookups fail, so do not try it sandboxed first and do not use `request_permissions` or `additional_permissions` for it.\n\nRun every helper command, from the first one, with these shell tool parameters:\n\n- `sandbox_permissions`: `\"require_escalated\"`\n- `justification`: one sentence saying the Atbash setup helper needs to reach Atbash and save its configuration outside the sandbox.\n- `prefix_rule`: `[\"node\", \"<skill-directory>/scripts/atbash-control.mjs\"]`, the same absolute launcher path as the command, so the user can approve the helper once instead of on every step.\n\nIf the shell tool has no `sandbox_permissions` parameter, Codex is already running without a sandbox: run the command as usual. If Codex rejects the escalation because its approval policy never asks, tell the user to switch Codex's permissions to Full access with `/permissions`, then run the same command again. A helper error that says Atbash \"could not be reached\" or \"could not write\" means the command ran inside the sandbox, not that the service is down: run the same command again once with the parameters above.\n\nRun the helper through this skill's `scripts/atbash-control.mjs` launcher, using the absolute skill directory:\n\n```text\nnode \"<skill-directory>/scripts/atbash-control.mjs\" setup start --host codex\n```\n\nUntil setup activates a profile, the hook allows only these setup steps, run outside the sandbox as above, and questions to the user, and denies everything else with \"Atbash is not set up yet\". Run each helper command as one plain command in exactly this form: no `cd`, `&&`, `;`, pipes, redirection, environment-variable prefixes, or command substitution, and no `--service` option. Commands in any other form are denied.\n\nThe result contains a public `verificationUri`, verification code, opaque job ID, and `planPath`. Show the URL and code to the user and ask them to complete wallet verification in **Connect Atbash**, then come back to the conversation. Never expose files under `~/.config/atbash/pending`, `credentials`, `profiles`, or `hosts`.\n\nInspect progress with `setup inspect <job-id>` through the same launcher. Follow `nextAction`:\n\n- `OPEN_BROWSER`: the user completes sign-in and wallet verification in the provided page.\n- `PREPARE_PLAN`: use discovery to determine what exists. Ask for any missing names or choices, then build a non-secret plan JSON containing only `actions` yourself and pass it to the launcher inline, as one single-quoted argument: `setup plan <job-id> --json '{\"actions\":[...]}'`. Do not write it to a file first — writing a file means running a shell command, which setup mode does not allow. Do not ask the user to write JSON or run the command.\n- If the organization has no active subscription, run `setup checkout <job-id> --organization <name>`. Show the returned `checkoutUrl` and ask the user to choose a plan in the dashboard's existing checkout flow. Resume inspection after the dashboard returns to `returnUrl`; never select paid plans or handle payment details yourself.\n- `REVIEW_IN_BROWSER`: the user reviews and signs the exact proposal in Connect Atbash. Do not approve it for them.\n- `WAIT`: inspect again after the returned poll interval, or after two seconds if none is present. Keep the user informed during longer waits; stop on expiry or a terminal failure.\n- `ACTIVATE`: run `setup continue <job-id>` to decrypt the delivered key locally and activate the profile.\n- `RECOVER`: report completed and failed steps. Any replacement mutation requires a new management session and approval.\n- `DONE`: `setup continue` already reports the new agent's `agentStatus`. Report setup as finished only when `nextAction` is `DONE` and `agentStatus.state` is `ready`. A session `status` of `completed` alone does not mean the profile is active. Do not run another command to check status: from this point every tool call is judged under the new agent's policy, and a blocked call can jail the agent. To confirm enforcement, make one harmless call that fits the agent's stated purpose.\n\nRegistering the hook is the user's step. If their `/hooks` list does not show the Atbash `PreToolUse` hook, ask them to run `node <skill-directory>/../../runtime/install-hook.cjs` from their own terminal, restart Codex, and trust the hook. Do not run the installer yourself: where the plugin's own hook already runs it registers a second copy, and once setup activates a profile it is judged like any other action. The plugin's bundled `hooks/hooks.json` entry runs a bare `node` and a `PLUGIN_ROOT` placeholder: it depends on `node` being on the host's PATH and on `PLUGIN_ROOT` being exported by the host, so it is not a gate on its own.\n\nFor a new setup, the plan normally contains `create_account` when missing, `create_organization` when missing, `activate_free_plan` when no subscription exists, then `create_agent` with `keySource: \"generate_in_browser\"`. Use only values the user supplied or explicitly chose. Do not invent organization names, organization descriptions, purposes, risks, or agent names; ask for any that are missing before building the plan.\n\n### Plan format\n\nA plan is `{\"actions\": [...]}` with the actions in the order below. There are no other top-level fields. Every action carries its own `\"network\"` (`\"public\"` or `\"private\"`), and all actions in one plan use the same network. Use a network listed in discovery's `availableNetworks`, which names the networks this Atbash deployment serves: when it lists only one, use it without asking; when it is absent, ask the user. Use exactly these fields:\n\n| Action | Include when | Fields |\n| --------------------- | ------------------------------------------ | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |\n| `create_account` | discovery shows no account on that network | `type`, `network` |\n| `create_organization` | the organization does not exist yet | `type`, `network`, `name` (2–64 letters, digits, `.`, `_`, or `-`), `description` (1–280 characters) |\n| `activate_free_plan` | the organization has no subscription | `type`, `network`, `organization` (the organization's name) |\n| `create_agent` | always, for a new agent | `type`, `network`, `organization`, `name` (up to 80 characters), `purpose` (up to 280), `risk` (`low`, `medium`, `high`, or `critical`), `keySource` (`\"generate_in_browser\"`), `policy` |\n\nFor example, a new user on a deployment that serves only the private network:\n\n```json\n{\n \"actions\": [\n { \"type\": \"create_account\", \"network\": \"private\" },\n {\n \"type\": \"create_organization\",\n \"network\": \"private\",\n \"name\": \"acme_ai\",\n \"description\": \"Acme agents\"\n },\n { \"type\": \"activate_free_plan\", \"network\": \"private\", \"organization\": \"acme_ai\" },\n {\n \"type\": \"create_agent\",\n \"network\": \"private\",\n \"organization\": \"acme_ai\",\n \"name\": \"codex dev\",\n \"purpose\": \"Coding assistant for the Acme repository\",\n \"risk\": \"medium\",\n \"keySource\": \"generate_in_browser\",\n \"policy\": { \"preset\": \"developer\", \"version\": 1 }\n }\n ]\n}\n```\n\nActivate the Free plan on the same network as the rest of the plan. If the dashboard answers that the Free plan is available only on the public network, that deployment keeps Free on public: a private organization there needs a paid plan first, so offer `setup checkout` for an existing organization or suggest the public network. Because the plan is passed in single quotes, write any single quote inside a value as `\\u0027` (for example `\"Nati\\u0027s agent\"`). If the helper rejects the plan, its error names the action number and the field to fix; correct that field and submit the whole plan again.\n\nGive every new agent a starting policy by adding `\"policy\": { \"preset\": \"<preset>\", \"version\": 1 }` to its `create_agent` action. Choose the preset from the agent's purpose: `developer` for coding work (the usual choice for Codex), `devops` for infrastructure, deployment, or operations work, and `full_baseline` when the user asks for the strictest protection. Tell the user which preset you suggested and that they can switch it or customize individual rules on the Connect Atbash review page before approving; the policy they approve is created as the organization policy \"<agent name> Policy\" and can be fine-tuned later under Policies in the dashboard. If the user wants a different preset before review, resubmit the plan with it.\n\n## Connect an existing agent\n\nAfter wallet verification and discovery, run `profile connect <job-id>` through the launcher. Show the returned loopback `localUri` to the user. If the user abandons the connection, or the local page expires before a key is entered, cancel the job as described under **Cancel setup**. The user enters the key in that local page. The helper derives its public key and connects it only if discovery shows the verified wallet owns the matching agent. The private key is never sent to Atbash or printed.\n\nNever ask the user to paste, upload, reveal, or dictate a private key. Never read or print credential files. Never put a key in a plan, prompt, environment assignment, command-line argument, committed file, or shell history.\n\n## Cancel setup\n\n`setup cancel <job-id>`, run through the same launcher, ends a setup job: it cancels the session in Connect Atbash, so its link stops working, and deletes the job's local files, including its session secret. Setup mode allows it, and like the other helper commands it runs outside the sandbox.\n\n- Run it when the user asks to stop, cancel, or start over, or abandons setup, while `nextAction` is `OPEN_BROWSER`, `PREPARE_PLAN`, or `REVIEW_IN_BROWSER`. Tell the user setup was cancelled. To start over, run `setup start` for a new job.\n- Run it at `RECOVER` when the session was cancelled, rejected, or expired, and for an expired job you will not resume, so its secrets do not stay on disk. If the helper reports that the session had already ended on Atbash, the cleanup still succeeded.\n- Do not cancel after the user approved the proposal: approved steps run in the browser and cancelling does not undo them. Let the job finish and report what was created. At `ACTIVATE` the new agent's key is waiting to be activated and the helper refuses to cancel; run `setup continue <job-id>` instead.\n\n## Profiles and legacy compatibility\n\nList profiles with `profile list --host codex`, select one with `profile switch --host codex --profile <id>`, or disconnect the host mapping with `profile disconnect --host codex`. Disconnecting retains the credential; it does not delete or revoke the on-chain agent.\n\nIf there is no selected profile, the hook keeps the legacy SDK configuration behavior. If `ATBASH_AGENT_KEY` or `ATBASH_ORG_NAME` conflicts with a selected profile, report the conflict and ask the user to remove or correct the override locally. Never inspect the conflicting key.\n\nSetup runs with the hook trusted; do not ask the user to disable the plugin, which also removes this skill. While no configuration exists at all, the hook allows the setup steps above and denies everything else.\n\nIf the hook denies a setup step because a configuration already exists (invalid, jailed, or not registered), setup mode does not apply — report the exact denial and tell the user they can disconnect the current profile from their own terminal with `node \"<skill-directory>/scripts/atbash-control.mjs\" profile disconnect --host codex`, then start setup again. Do not describe this as bypassing an individual verdict.\n\nIf the API fails, report the failing step and the returned error. A backend error is not a reason to hand the same command to the user or ask them to supply backend credentials. Do not repeatedly create sessions or claim that a manual command will fix a server failure.\n"Full snapshot data
{
"description": "Set up, connect, verify, troubleshoot, or switch the Atbash Safety profile for Codex. Use when the user wants to sign up for Atbash, connect an existing Atbash agent, create a new agent, check setup progress, or repair local plugin configuration.",
"included_files": [
{
"relative_path": "agents/openai.yaml",
"size_in_bytes": 183
},
{
"relative_path": "scripts/atbash-control.mjs",
"size_in_bytes": 551
},
{
"relative_path": "scripts/prepare-config.mjs",
"size_in_bytes": 2788
}
],
"name": "atbash-setup",
"skill_md_contents": "---\nname: atbash-setup\ndescription: Set up, connect, verify, troubleshoot, or switch the Atbash Safety profile for Codex. Use when the user wants to sign up for Atbash, connect an existing Atbash agent, create a new agent, check setup progress, or repair local plugin configuration.\n---\n\n# Atbash Setup\n\nUse the bundled control helper for onboarding. Keep private keys outside the conversation and outside tool arguments.\n\n## Carry out setup\n\nTreat a request to sign up, set up, or connect Atbash as a request to perform the workflow. Run the bundled commands yourself with the available execution tool; command examples below are for you to execute, not instructions to hand back to the user. Resolve `<skill-directory>` to the directory containing this SKILL.md, and use that installed copy's launcher throughout the job.\n\nHandle session creation, progress checks, discovery, plan-file creation and submission, and local activation yourself. Ask only for missing user choices and actions requiring the user's identity or consent: wallet sign-in, review and signature of the exact proposal, local entry of an existing agent key, and enabling or trusting the host hook. Do not send the user to manually create an account, organization, or agent when the control API supports that step.\n\nAfter a user completes verification or approval, inspect the same job and perform the next available step. Keep the job ID in the conversation so setup can resume without asking the user to run commands. Do not claim completion until the helper and status checks confirm it.\n\n## Start or resume setup\n\nThe helper runs outside the Codex sandbox. It must resolve and reach the Atbash dashboard, the chain nodes and its own loopback key page, and write the Atbash configuration directory (`~/.config/atbash`, or `$ATBASH_CONFIG_DIR` when it is set). Inside the sandbox, even with network access granted, its DNS lookups fail, so do not try it sandboxed first and do not use `request_permissions` or `additional_permissions` for it.\n\nRun every helper command, from the first one, with these shell tool parameters:\n\n- `sandbox_permissions`: `\"require_escalated\"`\n- `justification`: one sentence saying the Atbash setup helper needs to reach Atbash and save its configuration outside the sandbox.\n- `prefix_rule`: `[\"node\", \"<skill-directory>/scripts/atbash-control.mjs\"]`, the same absolute launcher path as the command, so the user can approve the helper once instead of on every step.\n\nIf the shell tool has no `sandbox_permissions` parameter, Codex is already running without a sandbox: run the command as usual. If Codex rejects the escalation because its approval policy never asks, tell the user to switch Codex's permissions to Full access with `/permissions`, then run the same command again. A helper error that says Atbash \"could not be reached\" or \"could not write\" means the command ran inside the sandbox, not that the service is down: run the same command again once with the parameters above.\n\nRun the helper through this skill's `scripts/atbash-control.mjs` launcher, using the absolute skill directory:\n\n```text\nnode \"<skill-directory>/scripts/atbash-control.mjs\" setup start --host codex\n```\n\nUntil setup activates a profile, the hook allows only these setup steps, run outside the sandbox as above, and questions to the user, and denies everything else with \"Atbash is not set up yet\". Run each helper command as one plain command in exactly this form: no `cd`, `&&`, `;`, pipes, redirection, environment-variable prefixes, or command substitution, and no `--service` option. Commands in any other form are denied.\n\nThe result contains a public `verificationUri`, verification code, opaque job ID, and `planPath`. Show the URL and code to the user and ask them to complete wallet verification in **Connect Atbash**, then come back to the conversation. Never expose files under `~/.config/atbash/pending`, `credentials`, `profiles`, or `hosts`.\n\nInspect progress with `setup inspect <job-id>` through the same launcher. Follow `nextAction`:\n\n- `OPEN_BROWSER`: the user completes sign-in and wallet verification in the provided page.\n- `PREPARE_PLAN`: use discovery to determine what exists. Ask for any missing names or choices, then build a non-secret plan JSON containing only `actions` yourself and pass it to the launcher inline, as one single-quoted argument: `setup plan <job-id> --json '{\"actions\":[...]}'`. Do not write it to a file first — writing a file means running a shell command, which setup mode does not allow. Do not ask the user to write JSON or run the command.\n- If the organization has no active subscription, run `setup checkout <job-id> --organization <name>`. Show the returned `checkoutUrl` and ask the user to choose a plan in the dashboard's existing checkout flow. Resume inspection after the dashboard returns to `returnUrl`; never select paid plans or handle payment details yourself.\n- `REVIEW_IN_BROWSER`: the user reviews and signs the exact proposal in Connect Atbash. Do not approve it for them.\n- `WAIT`: inspect again after the returned poll interval, or after two seconds if none is present. Keep the user informed during longer waits; stop on expiry or a terminal failure.\n- `ACTIVATE`: run `setup continue <job-id>` to decrypt the delivered key locally and activate the profile.\n- `RECOVER`: report completed and failed steps. Any replacement mutation requires a new management session and approval.\n- `DONE`: `setup continue` already reports the new agent's `agentStatus`. Report setup as finished only when `nextAction` is `DONE` and `agentStatus.state` is `ready`. A session `status` of `completed` alone does not mean the profile is active. Do not run another command to check status: from this point every tool call is judged under the new agent's policy, and a blocked call can jail the agent. To confirm enforcement, make one harmless call that fits the agent's stated purpose.\n\nRegistering the hook is the user's step. If their `/hooks` list does not show the Atbash `PreToolUse` hook, ask them to run `node <skill-directory>/../../runtime/install-hook.cjs` from their own terminal, restart Codex, and trust the hook. Do not run the installer yourself: where the plugin's own hook already runs it registers a second copy, and once setup activates a profile it is judged like any other action. The plugin's bundled `hooks/hooks.json` entry runs a bare `node` and a `PLUGIN_ROOT` placeholder: it depends on `node` being on the host's PATH and on `PLUGIN_ROOT` being exported by the host, so it is not a gate on its own.\n\nFor a new setup, the plan normally contains `create_account` when missing, `create_organization` when missing, `activate_free_plan` when no subscription exists, then `create_agent` with `keySource: \"generate_in_browser\"`. Use only values the user supplied or explicitly chose. Do not invent organization names, organization descriptions, purposes, risks, or agent names; ask for any that are missing before building the plan.\n\n### Plan format\n\nA plan is `{\"actions\": [...]}` with the actions in the order below. There are no other top-level fields. Every action carries its own `\"network\"` (`\"public\"` or `\"private\"`), and all actions in one plan use the same network. Use a network listed in discovery's `availableNetworks`, which names the networks this Atbash deployment serves: when it lists only one, use it without asking; when it is absent, ask the user. Use exactly these fields:\n\n| Action | Include when | Fields |\n| --------------------- | ------------------------------------------ | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |\n| `create_account` | discovery shows no account on that network | `type`, `network` |\n| `create_organization` | the organization does not exist yet | `type`, `network`, `name` (2–64 letters, digits, `.`, `_`, or `-`), `description` (1–280 characters) |\n| `activate_free_plan` | the organization has no subscription | `type`, `network`, `organization` (the organization's name) |\n| `create_agent` | always, for a new agent | `type`, `network`, `organization`, `name` (up to 80 characters), `purpose` (up to 280), `risk` (`low`, `medium`, `high`, or `critical`), `keySource` (`\"generate_in_browser\"`), `policy` |\n\nFor example, a new user on a deployment that serves only the private network:\n\n```json\n{\n \"actions\": [\n { \"type\": \"create_account\", \"network\": \"private\" },\n {\n \"type\": \"create_organization\",\n \"network\": \"private\",\n \"name\": \"acme_ai\",\n \"description\": \"Acme agents\"\n },\n { \"type\": \"activate_free_plan\", \"network\": \"private\", \"organization\": \"acme_ai\" },\n {\n \"type\": \"create_agent\",\n \"network\": \"private\",\n \"organization\": \"acme_ai\",\n \"name\": \"codex dev\",\n \"purpose\": \"Coding assistant for the Acme repository\",\n \"risk\": \"medium\",\n \"keySource\": \"generate_in_browser\",\n \"policy\": { \"preset\": \"developer\", \"version\": 1 }\n }\n ]\n}\n```\n\nActivate the Free plan on the same network as the rest of the plan. If the dashboard answers that the Free plan is available only on the public network, that deployment keeps Free on public: a private organization there needs a paid plan first, so offer `setup checkout` for an existing organization or suggest the public network. Because the plan is passed in single quotes, write any single quote inside a value as `\\u0027` (for example `\"Nati\\u0027s agent\"`). If the helper rejects the plan, its error names the action number and the field to fix; correct that field and submit the whole plan again.\n\nGive every new agent a starting policy by adding `\"policy\": { \"preset\": \"<preset>\", \"version\": 1 }` to its `create_agent` action. Choose the preset from the agent's purpose: `developer` for coding work (the usual choice for Codex), `devops` for infrastructure, deployment, or operations work, and `full_baseline` when the user asks for the strictest protection. Tell the user which preset you suggested and that they can switch it or customize individual rules on the Connect Atbash review page before approving; the policy they approve is created as the organization policy \"<agent name> Policy\" and can be fine-tuned later under Policies in the dashboard. If the user wants a different preset before review, resubmit the plan with it.\n\n## Connect an existing agent\n\nAfter wallet verification and discovery, run `profile connect <job-id>` through the launcher. Show the returned loopback `localUri` to the user. If the user abandons the connection, or the local page expires before a key is entered, cancel the job as described under **Cancel setup**. The user enters the key in that local page. The helper derives its public key and connects it only if discovery shows the verified wallet owns the matching agent. The private key is never sent to Atbash or printed.\n\nNever ask the user to paste, upload, reveal, or dictate a private key. Never read or print credential files. Never put a key in a plan, prompt, environment assignment, command-line argument, committed file, or shell history.\n\n## Cancel setup\n\n`setup cancel <job-id>`, run through the same launcher, ends a setup job: it cancels the session in Connect Atbash, so its link stops working, and deletes the job's local files, including its session secret. Setup mode allows it, and like the other helper commands it runs outside the sandbox.\n\n- Run it when the user asks to stop, cancel, or start over, or abandons setup, while `nextAction` is `OPEN_BROWSER`, `PREPARE_PLAN`, or `REVIEW_IN_BROWSER`. Tell the user setup was cancelled. To start over, run `setup start` for a new job.\n- Run it at `RECOVER` when the session was cancelled, rejected, or expired, and for an expired job you will not resume, so its secrets do not stay on disk. If the helper reports that the session had already ended on Atbash, the cleanup still succeeded.\n- Do not cancel after the user approved the proposal: approved steps run in the browser and cancelling does not undo them. Let the job finish and report what was created. At `ACTIVATE` the new agent's key is waiting to be activated and the helper refuses to cancel; run `setup continue <job-id>` instead.\n\n## Profiles and legacy compatibility\n\nList profiles with `profile list --host codex`, select one with `profile switch --host codex --profile <id>`, or disconnect the host mapping with `profile disconnect --host codex`. Disconnecting retains the credential; it does not delete or revoke the on-chain agent.\n\nIf there is no selected profile, the hook keeps the legacy SDK configuration behavior. If `ATBASH_AGENT_KEY` or `ATBASH_ORG_NAME` conflicts with a selected profile, report the conflict and ask the user to remove or correct the override locally. Never inspect the conflicting key.\n\nSetup runs with the hook trusted; do not ask the user to disable the plugin, which also removes this skill. While no configuration exists at all, the hook allows the setup steps above and denies everything else.\n\nIf the hook denies a setup step because a configuration already exists (invalid, jailed, or not registered), setup mode does not apply — report the exact denial and tell the user they can disconnect the current profile from their own terminal with `node \"<skill-directory>/scripts/atbash-control.mjs\" profile disconnect --host codex`, then start setup again. Do not describe this as bypassing an individual verdict.\n\nIf the API fails, report the failing step and the returned error. A backend error is not a reason to hand the same command to the user or ask them to supply backend credentials. Do not repeatedly create sessions or claim that a manual command will fix a server failure.\n"
}SHA-256 of public snapshot: 0599974a86b41cdda3cb34e0633a3f09989f4bc505241c8bda0ac7f83ea44952