tools
Registered agent tools.
| Name | Description | Source |
|---|---|---|
account_usage # | Inspect Codex account-wide quota percentages and reset times (not context capacity or conversation tokens). Demand refresh, at most once per minute; failures retain stale data. | extensions/adapters/providers/openai/usage.fnl:494 |
agent_state # | Read structured state of the running agent. Read-only; does not evaluate code. Query is a tiny Fennel-shaped data language. Examples: (:get :model), (:get :thinking), (:count (:get :messages)), (:get :messages -1), (:pluck (:get :tools) :name), (:get :extensions :panels), (:where (:get :messages) :role :assistant), (:last (:where (:get :messages) :role :assistant)), (:slice (:get :messages) -5 5), (:keys (:get)). Prefer narrow queries over dumping large roots. Output defaults to JSON; use format=fennel for Fennel rendering when available. | extensions/behaviors/companions/agent-state/init.fnl:94 |
builtin-tool-registry # | Dynamic loop registering every built-in tool spec from the builtin tools registry. | extensions/behaviors/kernel/builtin-tools/init.fnl:20 |
compact # | Compact this session's model context when it is becoming too large. Summarizes older messages, keeps recent messages verbatim, and persists the compaction for session resume. Call only when substantial context can be discarded; do not call repeatedly or on short sessions. | extensions/behaviors/companions/compact/init.fnl:415 |
extension # | Inspect loaded extensions and live registries, or reload one external extension. Actions: list, show, registry, reload. Read actions do not require interactive state; reload does. | extensions/behaviors/inspectors/extensions/commands/extension.fnl:631 |
fen_docs # | Read or search fen runtime docs and extension contracts. Useful for implementing extensions: inspect register kinds, canonical types, event shapes, and live commands/tools/providers. Topics: topics, commands, tools, providers, auth-backends, session-backends, presenters, controls, status, panels, prompt-fragments, introspectors, events, types, register-kinds, interfaces, extensions. Use name for a specific entry, e.g. {topic:'register-kinds', name:'tool'} or {topic:'types', name:'ToolResultMessage'}. Use query to search docs, optionally scoped by topic. | extensions/behaviors/kernel/docs/init.fnl:602 |
goal # | Start a bounded autonomous goal workflow during the active agent turn. | extensions/behaviors/companions/goal/init.fnl:780 |
models # | Inspect model providers without exposing credentials. Actions: current, providers, and list. | extensions/behaviors/companions/agent-state/init.fnl:111 |
plan # | Draft, revise, show, or cancel a read-only execution plan. This tool never approves or executes a plan. | extensions/behaviors/companions/plan/init.fnl:344 |
profile # | Control fen's statistical profiler for self-investigation. Actions: start, status, report, stop, reset, or save. Start accepts period (at least 100) and mode (functions or lines); tool save output-directory is confined to fen's profiles artifact root (or omit it to use an operator-configured default). Samples measure Lua VM instructions, not wall-clock time. | extensions/behaviors/inspectors/profiler-tool/init.fnl:8 |
queue # | Inspect pending steering and follow-up input. This agent-facing tool is read-only; use /queue for explicit user-owned clear and mode changes. | extensions/behaviors/inspectors/queue/commands/queue.fnl:178 |
reload # | Queue a reload request for after the current agent turn is fully idle. scope is reload for normal state-preserving reload, or registries for explicit registry recovery; reason is recorded for the user and logs. Restart and resume remain the full-process recovery path. | extensions/behaviors/actions/sessions/commands/session.fnl:489 |
simplify # | Queue a behavior-preserving simplification pass over changed code in the active agent turn. | extensions/behaviors/companions/simplify/init.fnl:245 |
skill # | Load a discovered skill's full instructions by name. The system prompt lists available skill names and descriptions; use this tool when a task matches one. | extensions/behaviors/companions/skills/init.fnl:717 |
subagent # | Use models before a launch to refresh exact models. review-worktrees creates 1–4 detached sibling worktrees for ordinary read-only subagent launches; cleanup-review-worktrees removes only unchanged worktrees it created. Other values inspect or manage runs. | extensions/behaviors/companions/subagent/init.fnl:2131 |
todo_write # | Create or update the structured todo list for this session. Use for non-trivial multi-step work. This tool overwrites the full current list; provide every item that should remain. Status must be pending, in_progress, or completed, with at most one in_progress item. Use an empty items array to clear the list. | extensions/behaviors/companions/todo/init.fnl:299 |