Repository navigation
skillhook 0.14.0
-
Respect Codex turn completion after recovered reconnect errors, preserving terminal failures and raw diagnostics.
-
Update Zod to the patched 4.6 line for GHSA-mrhx-c7xc-m3w5.
-
Each account's own MCP servers, connectors and plugins:
skillhook accounts show <name> --mcplists them with whether each connected, needs authentication or failed (and why), with tool counts, from the account's own CLI and environment: Claude Code'smcp_statuscontrol request (claude.ai connectors and plugins' servers included) and Codex'smcpServerStatus/listmerged withcodex mcp list --json.accounts --mcpadds a7/9column,accounts check --mcpfails on a server that needs authentication or fails,GET /accounts?mcp=1and MCPlist_accounts {mcp: true}return them, and the attention itemsmcp_needs_auth,mcp_failedandmcp_missing(servers the machine's own login has that the account lacks) carry the command that fixes each, run as the account.skillhook health(deep) has anaccount <name> mcp <server>row per server. Headers and environment values from the CLIs' answers are never kept. -
Several Claude Code and Codex logins on one machine:
skillhook accounts(docs/accounts.md). An account is a configuration directory of its own (CLAUDE_CONFIG_DIR/CODEX_HOME, under~/.skillhook/accounts/<name>; a Codex account links the machine's Codex settings exceptauth.json), a long-livedclaude setup-tokentoken or an API key.accounts add|login|logout|default|enable|disable|reset|discover|sync|exec|removemanage them;accounts,accounts showandaccounts usageshow for each who it is logged in as, its plan, its usage windows (5h, weekly, per model, credits; read through the CLI's own control protocol andcodex app-server, never with a token of skillhook's), its last use and its jobs of the last 24 hours, and what needs a person with the command that fixes it (logged out, login expired, usage limit, high usage, a failed login, a duplicate login).accounts check [--notify]exits 1 when one needs attention. -
account:in a skill or hook (anddefaults.accounts) runs it as an account, or a pool of accounts: each job runs as the first usable one (account_policy.strategy: ordered|balanced), moves to the next when a run fails with a usage limit or a failed login before the agent did anything, and skips a limited account until its limit resets (from the CLI's own notice or message). A run gets only its account's credentials. Jobs recordaccount(and each attempt's);skillhook run --account,POST /skills/<name>/run {account},jobs list --account,GET /jobs?account=. -
GET /accounts, the MCP toollist_accounts, anaccount <name>row per account inskillhook health, and the eventsaccount.changed/account.updated.servechecks every account at start, reads usage everyaccount_policy.usage_poll_minuteswhen set, and withaccount_policy.notify: desktopshows a notification when an account stops working or recovers. Readiness now tells an expired Claude Code login from none. -
Skillhook Cloud sees account health: a paired machine uploads
account.changed/account.updated(checked once when the link connects) and answers the read commandaccounts.get, through an allow-list: name, runner, kind, status, attention codes, plan, usage windows and reset times, last failure; emails masked toa…@example.com, no directories or paths,.envvalues scrubbed (docs/cloud.md).