Feature stages
hotl labels every user-facing switch with a stage. hotl features prints this
table; hotl doctor warns about each non-stable switch you have turned on.
| Stage | What it promises |
|---|---|
| experimental | May change or go without notice. Sandbox and egress escape hatches are labelled in every ask. |
| beta | Shipped and used; the shape may still move, and it will get a config twin. |
| stable | Covered by the distribution contract: a breaking change waits for a minor release and is in the changelog. |
| deprecated | Still works; the replacement is named; removal is announced one minor release ahead. |
Being listed changes nothing about how a switch behaves. 0.29.0 means the
next release.
| Switch | Stage | Since | What |
|---|---|---|---|
HOTL_MODEL |
stable | 0.2.0 | the session model (provider/model) |
HOTL_PROVIDER_BASE_URL |
stable | 0.12.0 | the active provider’s base URL |
HOTL_PROVIDER_AUTH |
stable | 0.4.0 | api_key or subscription |
HOTL_API_KEY_HELPER |
stable | 0.2.0 | a command whose stdout is the API key |
HOTL_API_KEY_HELPER_TTL_SECS |
stable | 0.2.0 | re-run the key helper after this long |
HOTL_CONTEXT_WINDOW |
stable | 0.2.0 | override the catalogued window |
HOTL_FAST_MODEL |
stable | 0.2.0 | housekeeping model |
HOTL_UTILITY_MODEL |
stable | 0.26.0 | model for digests and goal checks |
HOTL_EFFORT |
stable | 0.12.0 | reasoning depth |
HOTL_EVICT_TOKENS |
stable | 0.2.0 | spill tool results larger than this |
HOTL_PERMISSIONS |
stable | 0.2.0 | ask |
HOTL_PLAN |
stable | 0.9.0 | start in plan mode |
HOTL_SANDBOX |
stable | 0.2.0 | off disables the bash floor |
HOTL_MAX_TURNS |
stable | 0.5.1 | model steps per prompt |
HOTL_MAX_TOOL_CALLS |
stable | 0.26.0 | tool calls per session |
HOTL_MAX_COST_USD |
stable | 0.26.0 | USD per session |
HOTL_MAX_TOKENS |
stable | 0.14.0 | per-sample output tokens |
HOTL_CONCURRENCY_REQUESTS |
stable | 0.5.0 | concurrent provider requests |
HOTL_CONCURRENCY_AGENTS |
stable | 0.5.0 | concurrent sub-agents |
HOTL_CONCURRENCY_BLOCKING_THREADS |
stable | 0.5.0 | blocking-pool size |
HOTL_MOUSE |
stable | 0.5.2 | mouse capture in the console |
HOTL_SCRUB_ENV |
stable | 0.5.2 | extra env names scrubbed from children |
HOTL_SANDBOX_PROBE_DIR |
stable | 0.5.2 | where the sandbox probe writes |
HOTL_SHELL |
stable | 0.12.0 | the POSIX shell bash runs |
HOTL_PREAPPROVED |
stable | 0.2.0 | admin preapproved-rules file |
HOTL_THINKING |
beta | 0.5.2 | extended thinking on/off (config key pending) |
HOTL_SCRUB_ENV_STRICT |
beta | 0.5.2 | heuristic scrub of secret-looking env |
HOTL_COMPACTION_RESET |
beta | 0.2.0 | fresh-slate folds |
HOTL_HIDE_CONTEXT_PCT |
beta | 0.2.0 | keep fullness out of the turn context |
HOTL_PROGRAM_STATUS |
beta | 0.29.0 | terminal progress/status sequences |
HOTL_UNIX_SOCKETS |
experimental | 0.5.2 | let sandboxed bash reach unix sockets |
HOTL_MACOS_AUTOMATION |
experimental | 0.5.2 | let sandboxed bash drive Apple Events |
HOTL_WEB_ALLOW_METADATA |
experimental | 0.5.2 | let web_fetch reach metadata addresses |
HOTL_PROXY_AUTH |
experimental | 0.5.2 | egress proxy credential mode |
HOTL_CONCURRENCY_SUBPROCS |
experimental | 0.5.0 | reserved; parsed but not wired |
HOTL_CONCURRENCY_WORKER_THREADS |
experimental | 0.5.0 | reserved; parsed but not wired |
HOTL_SHELL_TRACE |
experimental | 0.29.0 | zsh plugin: attach the last command to : |
HOTL_TRACE_LINES |
experimental | 0.29.0 | zsh plugin: lines of output captured |
HOTL_TRACE_MAX_AGE |
experimental | 0.29.0 | zsh plugin: oldest trace attached |
HOTL_PROMPT_LINES |
experimental | 0.29.0 | zsh plugin: rows your prompt takes |
HOTL_OPENAI_BASE_URL |
deprecated | 0.2.0 | alias of the provider base URL — use HOTL_PROVIDER_BASE_URL |
HOTL_ANTHROPIC_BASE_URL |
deprecated | 0.4.0 | alias of the provider base URL — use HOTL_PROVIDER_BASE_URL |
HOTL_CACHE_WARMING / [provider.cache_warming] |
experimental | 0.29.0 | keep the 5-minute prompt cache alive while a tool runs |
[provider.cache_breakpoints] |
experimental | 0.24.0 | explicit GPT-5.6 prompt-cache breakpoints |
[agents.max_depth] |
experimental | 0.29.0 | how deep spawned agents may nest |
[agents.max_children] |
experimental | 0.29.0 | running plus parked sub-agents |
[behavior.max_jobs] |
experimental | 0.29.0 | running background jobs per session |