[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7AX3AEFDJHJJMVYTPF79P61":3,"skill-same:01M7AX3AEFDJHJJMVYTPF79P61":45},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":17,"publishedAt":18,"linkOnly":19,"markdownContent":20,"markdownBody":21,"frontMatter":22,"refName":23,"commitSha":24,"skillDirectory":25,"forksCount":26,"repositoryHtmlUrl":27,"repositoryHomepageUrl":6,"ownerHtmlUrl":28,"ownerBlogUrl":6,"ownerDescription":29,"updatedAt":30,"install":31,"resources":44},"01M7AX3AEFDJHJJMVYTPF79P61","theater-configure",null,"Configure or personalize Theater interactively when the user wants to set up, review, or change their Theater settings — favourite CLI, régie theme and behaviour, which models and thinking levels agents may use, plugin enablement, safety limits, retention, and observability. Use for a guided question-by-question walkthrough of config.toml that discovers what it can from the machine and asks the rest in plain language, including first-time setup on a new machine.","mana-byte\u002Ftheater","mana-byte","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F93316844?v=4","01M7AX1XSX6E0DJHN1SA4MEAJ3","01M7AX1XTJYZBKJH863Y7N1D61",1,"MIT","https:\u002F\u002Fgithub.com\u002Fmana-byte\u002Ftheater\u002Fblob\u002Fmain\u002Ftheater\u002Fskills\u002Fbuiltin\u002Ftheater-configure\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: theater-configure\ndescription: Configure or personalize Theater interactively when the user wants to set up, review, or change their Theater settings — favourite CLI, régie theme and behaviour, which models and thinking levels agents may use, plugin enablement, safety limits, retention, and observability. Use for a guided question-by-question walkthrough of config.toml that discovers what it can from the machine and asks the rest in plain language, including first-time setup on a new machine.\n---\n\n# Theater Configure\n\nRun a short, structured configuration interview and finish with a config.toml that contains\nexactly what the user chose and nothing else. Theater is configured per machine, one file,\nread at process start — `theater config path` prints where. Work in three movements: gather\nevery fact yourself, ask only for preferences, then write and validate.\n\n## Session rules\n\n- Use the host harness's native structured ask-user tool whenever one is available, giving\n  concrete options for every question whose answers are enumerable; otherwise ask in a normal\n  assistant response and stop until the user answers. Never invent an answer, and never write\n  a key the user did not choose.\n- Ask preference questions in plain, non-technical language — \"What's your favourite CLI?\",\n  \"What's your favourite theme?\" — and map the answers to config keys yourself. The user\n  never has to read or write TOML.\n- Facts are yours to find, never the user's. Installed CLIs, available models, available\n  plugins, and everything already in the file are discovered in step 1, not asked.\n- Every question carries a recommended answer, so a reply can be a single word.\n- A multi-select always offers a recommended combined option (\"all of them\", \"none\") so\n  the default is one click; when a question has more answers than the ask tool's option\n  limit, offer the recommended ones and let the rest arrive as free text.\n- A setting that already has a value from the user's file is reported, not re-asked: \"your\n  theme is nord — keep it?\" Only a requested change writes a key.\n- The user may say \"just the essentials\" or \"defaults everywhere\" at any point. Then drop\n  the remaining rounds, keep what is already answered, and go to the write-up.\n\n## Step 1 — Read the machine\n\nRun all of these before asking anything. They need no daemon running.\n\n- `theater config path` — where the file lives.\n- `theater config --json` — every resolved setting, tagged `config.toml` or `default`.\n  These are the values to present; never re-ask one the file already answers.\n- `theater harnesses --json` — which CLIs are installed, loadable, or broken.\n- `theater models --discover \u003Charness> --json` for vibe and opencode only — these two\n  CLIs answer discovery. The discovered block is written whole in step 3, spellings\n  untouched; it is what the user's CLIs are already configured to run.\n- `theater plugins --json` — local plugin packages; note the MCP ones. If there are none,\n  plugins are never mentioned again. Plugin packages live at `plugins\u002F\u003Cname>\u002F` under the\n  home, and a plugin may ship its own skills at `plugins\u002F\u003Cname>\u002Fskills\u002F\u003Cskill-name>\u002FSKILL.md` —\n  this layout is constant on every machine, so those relative paths can be read directly.\n\n### Models — the exact recipe per CLI\n\nAvailable models are facts. Enumerate them, never guess, and never settle for the\ndefault model — it is one entry, not the available set. pi, codex, and claude do not\nanswer discovery, so use these exact commands and record fields:\n\n**pi** — run `pi --list-models`. Each row is\n`provider  model  context  max-out  thinking  images`; take the first two columns only\nand write each row as one entry spelled `provider\u002Fmodel` (for example\n`mistral\u002Fzai-glm-5-2`) — that is the form pi's `--model` accepts. Every row is an\navailable model; do not filter. If the `pi` binary is missing, fall back to\n`~\u002F.pi\u002Fagent\u002Fmodels.json`: for every provider key in `providers`, every model under it\nbecomes one `provider\u002Fmodel` entry.\n\n**codex** — union three sources:\n1. `model = \"…\"` at the top of `~\u002F.codex\u002Fconfig.toml`.\n2. Every key of the `tui.model_availability_nux` table in that same file.\n3. The models actually served, from session records: for each file matching\n   `~\u002F.codex\u002Fsessions\u002F**\u002F*.jsonl`, parse lines until one has `\"type\": \"turn_context\"`,\n   take that record's `payload.model` once per file, then stop reading that file.\nSkip `codex-auto-review` — an internal alias, not a picker entry.\n\n**claude** — union two sources, then add aliases:\n1. `model` in `~\u002F.claude\u002Fsettings.json`.\n2. The models actually used: for each `~\u002F.claude\u002Fprojects\u002F**\u002F*.jsonl`, parse each line\n   that has a `message` object, take `message.model`, skip the literal `\u003Csynthetic>`,\n   keep distinct values only.\n3. Add `sonnet`, `opus`, `haiku` — the tier aliases `claude --model` accepts, which\n   track the newest version of each tier.\n\n**Two traps, for every CLI.** Take model names only from the record fields named above,\nnever from message content — prompts and tool output routinely contain model-looking\nstrings that were never available. And if a harness produced no entries at all, leave\nits key out and name it in the summary; never write a guess.\n\n## Step 2 — Interview, in rounds\n\nOne round at a time; later questions may depend on earlier answers. Multi-select where marked.\n\n### Round 1 — favourites\n\n1. `theater.favourite` — \"Which of these CLIs do you like best?\" Options: the installed\n   harnesses. Recommend the one this conversation runs in, or the first installed one.\n   Absent is a valid answer: no favourite.\n2. `regie.theme` — \"What's your favourite theme for the dashboard?\" Options:\n   ansi-dark, ansi-light, atom-one-dark, atom-one-light, catppuccin-frappe,\n   catppuccin-latte, catppuccin-macchiato, catppuccin-mocha, dracula, flexoki, gruvbox,\n   monokai, nord, rose-pine, rose-pine-dawn, rose-pine-moon, solarized-dark,\n   solarized-light, textual-dark, textual-light, tokyo-night. Offer a handful that suit\n   the user's terminal plus free text. Unset means Textual's own default, and an unknown\n   name is reported in the régie's status line rather than crashing — a miss is cosmetic.\n\n### Round 2 — grants and plugins\n\nModels and thinking levels are reproducible from the machine, so they are filled by\ndiscovery, never asked. Tell the user during this round, before anything is written, and\nagain in the closing summary, plainly, that a listed model or thinking level is a spending\nallowance — an agent that can spawn may use it without asking a human — so the user knows\nwhat was granted before the file takes effect.\n\n3. `[models]`, per harness — write the set the step 1 recipe produced for each\n   installed CLI, whole and unfiltered, in the CLI's own spelling (exact and\n   case-sensitive). Write a key only for a harness you actually filled.\n4. `[reasoning]`, per harness that takes a thinking level — write every level the CLI\n   advertises as of this release: codex — none, minimal, low, medium, high, xhigh, max,\n   ultra; claude — low, medium, high, xhigh, max; pi — the `--thinking \u003Clevel>` line of `pi --help`\n   lists exactly: off, minimal, low, medium, high, xhigh, max — run that one command and\n   copy the list verbatim, never from memory. opencode and vibe take none — never write\n   a key for them. Matching is exact.\n5. `[mcp] enabled` — only if MCP plugin packages were found: \"Do you want any of these\n   extras switched on?\" Options: the plugin names, multi-select. An explicit list replaces\n   the shipped default, and `beads` ships enabled — say so, and keep `beads` in a written\n   list unless the user explicitly declines it. Some plugins need\n   settings of their own before they will actually switch on — `theater plugins` shows\n   each one's state and error. For each chosen plugin that does, in order:\n   - Check whether the plugin ships a configure skill of its own —\n     `plugins\u002F\u003Cname>\u002Fskills\u002F` under the home; reading the directory needs no daemon.\n     If a configure skill is there, follow it to fill the plugin's\n     `[mcp.plugins.\u003Cname>]` table. Load it through your own skill-loading MCP tool\n     when its name appears in your skill listing; `theater skills` lists but does\n     not load. Otherwise read that `SKILL.md` file directly and follow it — this\n     includes the common case where a daemon is running but the plugin is disabled\n     or not yet configured, so its skill is absent from the listing; the direct read\n     is instructions for you only, and nothing enters the skill registry through it.\n   - If it ships none, ask plainly: \"This extra needs its own custom settings before it\n     will switch on — leave it off for now, or switch it on and you set it up yourself\n     later?\" Leave off: the name is not written into `enabled`. Switch on and DIY:\n     write it, and say in the summary that it stays dormant until the user fills its\n     `[mcp.plugins.\u003Cname>]` table by hand.\n6. `[skills] disabled` — \"Theater ships a few built-in helpers your agents can load —\n   any of them switched off?\" Options: the built-in skill names, multi-select,\n   recommend none; a bundled play counts as one option under its play name (`hierarchy`, `pair`),\n   since its `\u003Cplay>\u002F\u003Cmember>` skills are switched off together. Say in the question\n   that this configure helper switches itself off\n   at the end of the session by design, so \"none\" still ends with it disabled — the\n   user who wants it kept can say so, and that counts as asking to keep it enabled.\n   The names come from `theater skills` (needs a daemon) or your own\n   skill-listing MCP tool; if neither is reachable, skip the question and write\n   nothing. A name in the list is simply not offered to agents. Disabling is presence,\n   not refusal — built-in skills stay validated, and an unknown name in the list is\n   not an error. A disabled skill also cannot be attached with `skills=[...]`: a\n   spawn or send naming it is refused.\n7. `[harness] disabled` — never asked; every CLI is driven by default. Only when\n   `theater harnesses` shows a broken plugin, recommend disabling it by its folder name\n   and write the denylist entry if the user agrees. Disabling is absence, not refusal,\n   and the name is matched against the plugin folder before import.\n\n### Round 3 — limits\n\n8. `rails.depth_cap` — \"How many levels of delegation should be allowed?\" A boss agent may\n   hire a helper, who may hire a helper… Default 3.\n9. `rails.budget` — \"How many agents may share one boss before Theater says enough?\"\n   Default 20. Raise deliberately; these two are what stops a runaway agent.\n10. `retention.jobs_days` — \"How long should Theater remember what past jobs did?\"\n    Default two weeks. The automatic database tidy-up itself is never asked about — it\n    stays on.\n\n### Round 4 — the régie's feel\n\n11. `regie.cost_window` — \"The footer can total what agents have cost you — over today,\n    this week, this month, or the whole year?\" Options: day (default), week, month, year.\n12. `regie.participant_detail` — \"In the sidebar, should each agent show its folder or\n    its description?\" Options: cwd (default), description.\n13. `regie.sidebar_width` — \"How wide should the sidebar be, in characters?\" Default 52;\n    below 40 the tree rails no longer fit.\n\n### Round 5 — observability\n\n14. The observability gate — \"Do you watch Theater in a telemetry dashboard (Grafana,\n    Jaeger, an OpenTelemetry collector)? Turning it on exports Theater's traces, metrics,\n    and logs to your collector, and needs the observability extra installed.\" Recommend\n    no; with no, the export stays off and no key is written. If yes: turn the export on,\n    ask for the collector address and whether it speaks grpc (default) or http, keep the\n    service name \"theater\" unless the user cares, and ask the same way about the\n    agent-signal toggles, the content opt-in, and the export and gauge intervals.\n\n### Silently defaulted — never asked, never written\n\nEverything else is a measured default that only a symptom or a curious user should\nsurface. Do not ask about it and do not write it; if the user names one, take the change\nand write only that key. The surface, so you can answer a follow-up without re-reading\nanything: the event log hidden on open; welcome animations on; built-in welcome\nsentences; two folder segments per agent; tree refresh 1s; event-log poll 0.4s in\nbatches of 50; ledger page 30; sentence and tip typing pacing; every observer timer\n(poll 0.25s, search 2s, relocate 5s, awaiting-input 1.5s, screen 1s, rescue 60s, sync\n1s); the tidy-up hourly in batches of 5000, event feed kept 7 days, refused-send cap\n10000, stale jobs 7 days; log files rolling at 10 MiB keeping 3 backups.\n\n## Step 3 — Write the file\n\n- Read the existing config.toml first, if there is one. Preserve every key, value, and\n  comment the interview did not change; edit surgically, never rewrite the file wholesale.\n  A fresh file contains only the chosen keys, each in its section.\n- Write only what the user chose or what discovery filled. An unwritten key keeps tracking\n  the code's default; a written one is pinned against future changes — so a setting kept at\n  its default is left out of the file, not written down.\n- Shapes: `favourite` and `theme` are strings; `[models]` and `[reasoning]` hold one list\n  per harness; `[mcp] enabled`, `[skills] disabled`, and `[harness] disabled` are lists.\n  Model and effort spellings are the CLI's own, exact and case-sensitive.\n- `[mcp] enabled` is authoritative: it replaces the shipped default, where `beads` is\n  enabled — so keep `beads` in a written list unless the user declined it, and if the\n  list already holds names, append — never drop one.\n- Last write of the session: add `theater-configure` itself to `[skills] disabled`.\n  The skill's work is done once the file validates, and a disabled configure skill\n  cannot be loaded by accident later. This is the one deliberate exception to writing\n  only what the user chose — it was announced in the interview's question 6, and the\n  user overrules it by asking to keep the skill enabled. If the list already holds\n  names, append — never drop one.\n- Never write a secret into the file. An MCP plugin's secrets belong in its\n  `[mcp.plugins.\u003Cname>]` table as `{ env = \"NAME\" }` or `{ file = \"\u002Fpath\" }`; filling\n  that table is the plugin's own configure skill's job when it ships one, otherwise\n  the user's.\n- Validate: `theater config` reports a malformed file exactly as the daemon would reject\n  it, and `theater models` shows the resulting allowlists. Fix until both are clean.\n  The last line of `theater config` says whether the running daemon needs\n  `theater reload` or `theater restart` to pick the file up.\n\n## Step 4 — Close\n\nEach Theater process reads the file at its own start; the daemon also re-reads it on\n`theater reload`. Offer `theater reload` first: it applies `[rails]`, `[models]`,\n`[reasoning]` and `[skills] disabled` in place and never starts a daemon. When it\nnames settings only a restart applies, it has applied nothing — then offer\n`theater restart`, which restarts only the daemon. Two things neither touches: a régie\nthat is already open picks up config-dependent visuals (theme, sidebar width) only when it is\nclosed and relaunched, and agents that are already running keep their old telemetry\nexport settings until they are relaunched, because each agent's MCP process configures\nobservability from the file at startup. Summarize what was discovered, what was written\nwhere, what was deliberately left absent (grants and pinned defaults), and what stays\nat its default. Say plainly that theater-configure switched itself off in\n`[skills] disabled`, and that undoing it is removing the name from that list and\nrunning `theater reload`. Do not create a report file.\n","# Theater Configure\n\nRun a short, structured configuration interview and finish with a config.toml that contains\nexactly what the user chose and nothing else. Theater is configured per machine, one file,\nread at process start — `theater config path` prints where. Work in three movements: gather\nevery fact yourself, ask only for preferences, then write and validate.\n\n## Session rules\n\n- Use the host harness's native structured ask-user tool whenever one is available, giving\n  concrete options for every question whose answers are enumerable; otherwise ask in a normal\n  assistant response and stop until the user answers. Never invent an answer, and never write\n  a key the user did not choose.\n- Ask preference questions in plain, non-technical language — \"What's your favourite CLI?\",\n  \"What's your favourite theme?\" — and map the answers to config keys yourself. The user\n  never has to read or write TOML.\n- Facts are yours to find, never the user's. Installed CLIs, available models, available\n  plugins, and everything already in the file are discovered in step 1, not asked.\n- Every question carries a recommended answer, so a reply can be a single word.\n- A multi-select always offers a recommended combined option (\"all of them\", \"none\") so\n  the default is one click; when a question has more answers than the ask tool's option\n  limit, offer the recommended ones and let the rest arrive as free text.\n- A setting that already has a value from the user's file is reported, not re-asked: \"your\n  theme is nord — keep it?\" Only a requested change writes a key.\n- The user may say \"just the essentials\" or \"defaults everywhere\" at any point. Then drop\n  the remaining rounds, keep what is already answered, and go to the write-up.\n\n## Step 1 — Read the machine\n\nRun all of these before asking anything. They need no daemon running.\n\n- `theater config path` — where the file lives.\n- `theater config --json` — every resolved setting, tagged `config.toml` or `default`.\n  These are the values to present; never re-ask one the file already answers.\n- `theater harnesses --json` — which CLIs are installed, loadable, or broken.\n- `theater models --discover \u003Charness> --json` for vibe and opencode only — these two\n  CLIs answer discovery. The discovered block is written whole in step 3, spellings\n  untouched; it is what the user's CLIs are already configured to run.\n- `theater plugins --json` — local plugin packages; note the MCP ones. If there are none,\n  plugins are never mentioned again. Plugin packages live at `plugins\u002F\u003Cname>\u002F` under the\n  home, and a plugin may ship its own skills at `plugins\u002F\u003Cname>\u002Fskills\u002F\u003Cskill-name>\u002FSKILL.md` —\n  this layout is constant on every machine, so those relative paths can be read directly.\n\n### Models — the exact recipe per CLI\n\nAvailable models are facts. Enumerate them, never guess, and never settle for the\ndefault model — it is one entry, not the available set. pi, codex, and claude do not\nanswer discovery, so use these exact commands and record fields:\n\n**pi** — run `pi --list-models`. Each row is\n`provider  model  context  max-out  thinking  images`; take the first two columns only\nand write each row as one entry spelled `provider\u002Fmodel` (for example\n`mistral\u002Fzai-glm-5-2`) — that is the form pi's `--model` accepts. Every row is an\navailable model; do not filter. If the `pi` binary is missing, fall back to\n`~\u002F.pi\u002Fagent\u002Fmodels.json`: for every provider key in `providers`, every model under it\nbecomes one `provider\u002Fmodel` entry.\n\n**codex** — union three sources:\n1. `model = \"…\"` at the top of `~\u002F.codex\u002Fconfig.toml`.\n2. Every key of the `tui.model_availability_nux` table in that same file.\n3. The models actually served, from session records: for each file matching\n   `~\u002F.codex\u002Fsessions\u002F**\u002F*.jsonl`, parse lines until one has `\"type\": \"turn_context\"`,\n   take that record's `payload.model` once per file, then stop reading that file.\nSkip `codex-auto-review` — an internal alias, not a picker entry.\n\n**claude** — union two sources, then add aliases:\n1. `model` in `~\u002F.claude\u002Fsettings.json`.\n2. The models actually used: for each `~\u002F.claude\u002Fprojects\u002F**\u002F*.jsonl`, parse each line\n   that has a `message` object, take `message.model`, skip the literal `\u003Csynthetic>`,\n   keep distinct values only.\n3. Add `sonnet`, `opus`, `haiku` — the tier aliases `claude --model` accepts, which\n   track the newest version of each tier.\n\n**Two traps, for every CLI.** Take model names only from the record fields named above,\nnever from message content — prompts and tool output routinely contain model-looking\nstrings that were never available. And if a harness produced no entries at all, leave\nits key out and name it in the summary; never write a guess.\n\n## Step 2 — Interview, in rounds\n\nOne round at a time; later questions may depend on earlier answers. Multi-select where marked.\n\n### Round 1 — favourites\n\n1. `theater.favourite` — \"Which of these CLIs do you like best?\" Options: the installed\n   harnesses. Recommend the one this conversation runs in, or the first installed one.\n   Absent is a valid answer: no favourite.\n2. `regie.theme` — \"What's your favourite theme for the dashboard?\" Options:\n   ansi-dark, ansi-light, atom-one-dark, atom-one-light, catppuccin-frappe,\n   catppuccin-latte, catppuccin-macchiato, catppuccin-mocha, dracula, flexoki, gruvbox,\n   monokai, nord, rose-pine, rose-pine-dawn, rose-pine-moon, solarized-dark,\n   solarized-light, textual-dark, textual-light, tokyo-night. Offer a handful that suit\n   the user's terminal plus free text. Unset means Textual's own default, and an unknown\n   name is reported in the régie's status line rather than crashing — a miss is cosmetic.\n\n### Round 2 — grants and plugins\n\nModels and thinking levels are reproducible from the machine, so they are filled by\ndiscovery, never asked. Tell the user during this round, before anything is written, and\nagain in the closing summary, plainly, that a listed model or thinking level is a spending\nallowance — an agent that can spawn may use it without asking a human — so the user knows\nwhat was granted before the file takes effect.\n\n3. `[models]`, per harness — write the set the step 1 recipe produced for each\n   installed CLI, whole and unfiltered, in the CLI's own spelling (exact and\n   case-sensitive). Write a key only for a harness you actually filled.\n4. `[reasoning]`, per harness that takes a thinking level — write every level the CLI\n   advertises as of this release: codex — none, minimal, low, medium, high, xhigh, max,\n   ultra; claude — low, medium, high, xhigh, max; pi — the `--thinking \u003Clevel>` line of `pi --help`\n   lists exactly: off, minimal, low, medium, high, xhigh, max — run that one command and\n   copy the list verbatim, never from memory. opencode and vibe take none — never write\n   a key for them. Matching is exact.\n5. `[mcp] enabled` — only if MCP plugin packages were found: \"Do you want any of these\n   extras switched on?\" Options: the plugin names, multi-select. An explicit list replaces\n   the shipped default, and `beads` ships enabled — say so, and keep `beads` in a written\n   list unless the user explicitly declines it. Some plugins need\n   settings of their own before they will actually switch on — `theater plugins` shows\n   each one's state and error. For each chosen plugin that does, in order:\n   - Check whether the plugin ships a configure skill of its own —\n     `plugins\u002F\u003Cname>\u002Fskills\u002F` under the home; reading the directory needs no daemon.\n     If a configure skill is there, follow it to fill the plugin's\n     `[mcp.plugins.\u003Cname>]` table. Load it through your own skill-loading MCP tool\n     when its name appears in your skill listing; `theater skills` lists but does\n     not load. Otherwise read that `SKILL.md` file directly and follow it — this\n     includes the common case where a daemon is running but the plugin is disabled\n     or not yet configured, so its skill is absent from the listing; the direct read\n     is instructions for you only, and nothing enters the skill registry through it.\n   - If it ships none, ask plainly: \"This extra needs its own custom settings before it\n     will switch on — leave it off for now, or switch it on and you set it up yourself\n     later?\" Leave off: the name is not written into `enabled`. Switch on and DIY:\n     write it, and say in the summary that it stays dormant until the user fills its\n     `[mcp.plugins.\u003Cname>]` table by hand.\n6. `[skills] disabled` — \"Theater ships a few built-in helpers your agents can load —\n   any of them switched off?\" Options: the built-in skill names, multi-select,\n   recommend none; a bundled play counts as one option under its play name (`hierarchy`, `pair`),\n   since its `\u003Cplay>\u002F\u003Cmember>` skills are switched off together. Say in the question\n   that this configure helper switches itself off\n   at the end of the session by design, so \"none\" still ends with it disabled — the\n   user who wants it kept can say so, and that counts as asking to keep it enabled.\n   The names come from `theater skills` (needs a daemon) or your own\n   skill-listing MCP tool; if neither is reachable, skip the question and write\n   nothing. A name in the list is simply not offered to agents. Disabling is presence,\n   not refusal — built-in skills stay validated, and an unknown name in the list is\n   not an error. A disabled skill also cannot be attached with `skills=[...]`: a\n   spawn or send naming it is refused.\n7. `[harness] disabled` — never asked; every CLI is driven by default. Only when\n   `theater harnesses` shows a broken plugin, recommend disabling it by its folder name\n   and write the denylist entry if the user agrees. Disabling is absence, not refusal,\n   and the name is matched against the plugin folder before import.\n\n### Round 3 — limits\n\n8. `rails.depth_cap` — \"How many levels of delegation should be allowed?\" A boss agent may\n   hire a helper, who may hire a helper… Default 3.\n9. `rails.budget` — \"How many agents may share one boss before Theater says enough?\"\n   Default 20. Raise deliberately; these two are what stops a runaway agent.\n10. `retention.jobs_days` — \"How long should Theater remember what past jobs did?\"\n    Default two weeks. The automatic database tidy-up itself is never asked about — it\n    stays on.\n\n### Round 4 — the régie's feel\n\n11. `regie.cost_window` — \"The footer can total what agents have cost you — over today,\n    this week, this month, or the whole year?\" Options: day (default), week, month, year.\n12. `regie.participant_detail` — \"In the sidebar, should each agent show its folder or\n    its description?\" Options: cwd (default), description.\n13. `regie.sidebar_width` — \"How wide should the sidebar be, in characters?\" Default 52;\n    below 40 the tree rails no longer fit.\n\n### Round 5 — observability\n\n14. The observability gate — \"Do you watch Theater in a telemetry dashboard (Grafana,\n    Jaeger, an OpenTelemetry collector)? Turning it on exports Theater's traces, metrics,\n    and logs to your collector, and needs the observability extra installed.\" Recommend\n    no; with no, the export stays off and no key is written. If yes: turn the export on,\n    ask for the collector address and whether it speaks grpc (default) or http, keep the\n    service name \"theater\" unless the user cares, and ask the same way about the\n    agent-signal toggles, the content opt-in, and the export and gauge intervals.\n\n### Silently defaulted — never asked, never written\n\nEverything else is a measured default that only a symptom or a curious user should\nsurface. Do not ask about it and do not write it; if the user names one, take the change\nand write only that key. The surface, so you can answer a follow-up without re-reading\nanything: the event log hidden on open; welcome animations on; built-in welcome\nsentences; two folder segments per agent; tree refresh 1s; event-log poll 0.4s in\nbatches of 50; ledger page 30; sentence and tip typing pacing; every observer timer\n(poll 0.25s, search 2s, relocate 5s, awaiting-input 1.5s, screen 1s, rescue 60s, sync\n1s); the tidy-up hourly in batches of 5000, event feed kept 7 days, refused-send cap\n10000, stale jobs 7 days; log files rolling at 10 MiB keeping 3 backups.\n\n## Step 3 — Write the file\n\n- Read the existing config.toml first, if there is one. Preserve every key, value, and\n  comment the interview did not change; edit surgically, never rewrite the file wholesale.\n  A fresh file contains only the chosen keys, each in its section.\n- Write only what the user chose or what discovery filled. An unwritten key keeps tracking\n  the code's default; a written one is pinned against future changes — so a setting kept at\n  its default is left out of the file, not written down.\n- Shapes: `favourite` and `theme` are strings; `[models]` and `[reasoning]` hold one list\n  per harness; `[mcp] enabled`, `[skills] disabled`, and `[harness] disabled` are lists.\n  Model and effort spellings are the CLI's own, exact and case-sensitive.\n- `[mcp] enabled` is authoritative: it replaces the shipped default, where `beads` is\n  enabled — so keep `beads` in a written list unless the user declined it, and if the\n  list already holds names, append — never drop one.\n- Last write of the session: add `theater-configure` itself to `[skills] disabled`.\n  The skill's work is done once the file validates, and a disabled configure skill\n  cannot be loaded by accident later. This is the one deliberate exception to writing\n  only what the user chose — it was announced in the interview's question 6, and the\n  user overrules it by asking to keep the skill enabled. If the list already holds\n  names, append — never drop one.\n- Never write a secret into the file. An MCP plugin's secrets belong in its\n  `[mcp.plugins.\u003Cname>]` table as `{ env = \"NAME\" }` or `{ file = \"\u002Fpath\" }`; filling\n  that table is the plugin's own configure skill's job when it ships one, otherwise\n  the user's.\n- Validate: `theater config` reports a malformed file exactly as the daemon would reject\n  it, and `theater models` shows the resulting allowlists. Fix until both are clean.\n  The last line of `theater config` says whether the running daemon needs\n  `theater reload` or `theater restart` to pick the file up.\n\n## Step 4 — Close\n\nEach Theater process reads the file at its own start; the daemon also re-reads it on\n`theater reload`. Offer `theater reload` first: it applies `[rails]`, `[models]`,\n`[reasoning]` and `[skills] disabled` in place and never starts a daemon. When it\nnames settings only a restart applies, it has applied nothing — then offer\n`theater restart`, which restarts only the daemon. Two things neither touches: a régie\nthat is already open picks up config-dependent visuals (theme, sidebar width) only when it is\nclosed and relaunched, and agents that are already running keep their old telemetry\nexport settings until they are relaunched, because each agent's MCP process configures\nobservability from the file at startup. Summarize what was discovered, what was written\nwhere, what was deliberately left absent (grants and pinned defaults), and what stays\nat its default. Say plainly that theater-configure switched itself off in\n`[skills] disabled`, and that undoing it is removing the name from that list and\nrunning `theater reload`. Do not create a report file.\n",{"name":5,"description":7},"main","8ff9aa9ce8393ccad496ed8806324b7b4b6c2314","theater\u002Fskills\u002Fbuiltin\u002Ftheater-configure",0,"https:\u002F\u002Fgithub.com\u002Fmana-byte\u002Ftheater","https:\u002F\u002Fgithub.com\u002Fmana-byte","Always learning","2026-11-14 20:40:25",{"source":8,"skillName":5,"commands":32,"prompt":42,"zipAvailable":43},[33,36,39],{"tool":34,"command":35},"npx","npx skills add mana-byte\u002Ftheater --skill theater-configure",{"tool":37,"command":38},"bunx","bunx skills add mana-byte\u002Ftheater --skill theater-configure",{"tool":40,"command":41},"pnpm","pnpm dlx skills add mana-byte\u002Ftheater --skill theater-configure","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fmana-byte\u002Ftheater\u002Fblob\u002Fmain\u002Ftheater\u002Fskills\u002Fbuiltin\u002Ftheater-configure\u002FSKILL.md",true,[],[46,53,60,67,74,81],{"skillNo":47,"name":48,"nameZh":6,"description":49,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":50,"categories":51,"keywords":52,"publishedAt":18},"01M7AX3QM639GSJHJAAE6VCAT1","theater-recover-tmux","Recover Theater participants after a tmux server restart or kill-server using retained participant metadata, native session resume, and an explicitly chosen conservative recovery mode.","https:\u002F\u002Fgithub.com\u002Fmana-byte\u002Ftheater\u002Fblob\u002Fmain\u002Ftheater\u002Fskills\u002Fbuiltin\u002Ftheater-recover-tmux\u002FSKILL.md",[],[],{"skillNo":54,"name":55,"nameZh":6,"description":56,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":57,"categories":58,"keywords":59,"publishedAt":18},"01M7AX3M8JZ8FPRCP4F3CFWDN7","theater-delegate","Hand a scoped task to a suitable existing Theater peer, or start a builder and independent reviewer pair when the work needs one. Stay or run in the background as the user requested.","https:\u002F\u002Fgithub.com\u002Fmana-byte\u002Ftheater\u002Fblob\u002Fmain\u002Ftheater\u002Fskills\u002Fbuiltin\u002Ftheater-delegate\u002FSKILL.md",[],[],{"skillNo":61,"name":62,"nameZh":6,"description":63,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":64,"categories":65,"keywords":66,"publishedAt":18},"01M7AX3FDEF136J646DZWVW3K6","theater-debate","Debate a review, design, or architecture decision with one specified Theater participant when the user asks to debate with a harness, model, and reasoning effort. Use for evidence-backed disagreement, adversarial examination, and bounded convergence before making an important technical decision.","https:\u002F\u002Fgithub.com\u002Fmana-byte\u002Ftheater\u002Fblob\u002Fmain\u002Ftheater\u002Fskills\u002Fbuiltin\u002Ftheater-debate\u002FSKILL.md",[],[],{"skillNo":68,"name":69,"nameZh":6,"description":70,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":71,"categories":72,"keywords":73,"publishedAt":18},"01M7AX36NADMJ5ZCAQC5T05ZM8","reviewer","Master of a pair play. Spawn one builder, review its work read-only, debate with evidence until you both agree, verify, and report. Attach this card when spawning a reviewer; never load it in your own chat.","https:\u002F\u002Fgithub.com\u002Fmana-byte\u002Ftheater\u002Fblob\u002Fmain\u002Ftheater\u002Fskills\u002Fbuiltin\u002Fpair\u002Freviewer\u002FSKILL.md",[],[],{"skillNo":75,"name":76,"nameZh":6,"description":77,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":78,"categories":79,"keywords":80,"publishedAt":18},"01M7AX34K85MAFMSA3CV0X8RW2","builder","Builder in a pair play. Implement the brief in the shared worktree, defend your choices with evidence, and iterate with the reviewer until you both agree.","https:\u002F\u002Fgithub.com\u002Fmana-byte\u002Ftheater\u002Fblob\u002Fmain\u002Ftheater\u002Fskills\u002Fbuiltin\u002Fpair\u002Fbuilder\u002FSKILL.md",[],[],{"skillNo":82,"name":83,"nameZh":6,"description":84,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":85,"categories":86,"keywords":87,"publishedAt":18},"01M7AX32XR0X6XDRXVK7FKYYN1","worker","Task role for a lead or member in a hierarchy play. Implement the assigned change inside the paths you own and report what you verified.","https:\u002F\u002Fgithub.com\u002Fmana-byte\u002Ftheater\u002Fblob\u002Fmain\u002Ftheater\u002Fmcp_plugins\u002Fbuiltin\u002Fbeads\u002Fskills\u002Fhierarchy\u002Fworker\u002FSKILL.md",[],[]]