docks-kit 0.20.5 → 0.20.8
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/AGENTS.md +19 -17
- package/README.md +17 -6
- package/cli/docs/command-guards.md +269 -0
- package/cli/docs/install.md +23 -12
- package/cli/docs/models.md +3 -2
- package/cli/docs/modifiers.md +4 -3
- package/cli/docs/omp-context.md +9 -2
- package/cli/docs/omp-models.md +61 -31
- package/cli/docs/plugins.md +12 -7
- package/cli/docs/sync-layers.md +26 -9
- package/cli/docs/toolchain.md +7 -4
- package/cli/src/argv.ts +2 -0
- package/cli/src/argvSurface.ts +23 -0
- package/cli/src/argvValidate.ts +3 -3
- package/cli/src/commands/docs.ts +6 -1
- package/cli/src/commands/harnesses.ts +15 -4
- package/cli/src/commands/omp.ts +92 -41
- package/cli/src/commands/status.ts +15 -0
- package/cli/src/commands/sync.ts +6 -6
- package/cli/src/commands/toolchain.ts +8 -2
- package/cli/src/commands/update.ts +51 -10
- package/cli/src/engine-native/DESIGN.md +56 -35
- package/cli/src/engine-native/atomicWrite.ts +64 -0
- package/cli/src/engine-native/claudeLsp.ts +9 -2
- package/cli/src/engine-native/claudeOptionalPlugins.ts +19 -6
- package/cli/src/engine-native/claudePluginPasses.ts +77 -21
- package/cli/src/engine-native/claudeRemovals.ts +245 -57
- package/cli/src/engine-native/claudeRetired.ts +1 -1
- package/cli/src/engine-native/claudeSettings.ts +46 -19
- package/cli/src/engine-native/claudeSettingsModifiers.ts +3 -3
- package/cli/src/engine-native/claudeSync.ts +18 -16
- package/cli/src/engine-native/codexConfig.ts +72 -12
- package/cli/src/engine-native/codexHooks.ts +3 -3
- package/cli/src/engine-native/codexPaths.ts +14 -0
- package/cli/src/engine-native/codexPlugins.ts +48 -23
- package/cli/src/engine-native/codexStatus.ts +2 -1
- package/cli/src/engine-native/codexSync.ts +33 -7
- package/cli/src/engine-native/codexToml.ts +161 -35
- package/cli/src/engine-native/deps.ts +4 -9
- package/cli/src/engine-native/engineCtx.ts +42 -14
- package/cli/src/engine-native/exec.ts +28 -1
- package/cli/src/engine-native/harnesses.ts +32 -8
- package/cli/src/engine-native/index.ts +6 -1
- package/cli/src/engine-native/kitDb.ts +12 -6
- package/cli/src/engine-native/liveModels.ts +2 -1
- package/cli/src/engine-native/models.ts +6 -3
- package/cli/src/engine-native/modes.ts +15 -10
- package/cli/src/engine-native/ompFileDeploy.ts +3 -4
- package/cli/src/engine-native/ompMarketplace.ts +9 -7
- package/cli/src/engine-native/ompOverlay.ts +39 -13
- package/cli/src/engine-native/ompPaths.ts +20 -9
- package/cli/src/engine-native/ompPlugins.ts +16 -15
- package/cli/src/engine-native/ompSync.ts +11 -6
- package/cli/src/engine-native/ompYaml.ts +91 -6
- package/cli/src/engine-native/os/darwin.ts +2 -0
- package/cli/src/engine-native/os/linux.ts +1 -0
- package/cli/src/engine-native/os/posix.ts +17 -5
- package/cli/src/engine-native/os/types.ts +2 -2
- package/cli/src/engine-native/os/windows.ts +24 -16
- package/cli/src/engine-native/skillsInstall.ts +10 -5
- package/cli/src/engine-native/skillsLinks.ts +14 -1
- package/cli/src/engine-native/skillsManifest.ts +20 -0
- package/cli/src/engine-native/skillsPrune.ts +33 -26
- package/cli/src/engine-native/skillsSync.ts +6 -3
- package/cli/src/engine-native/syncDispatch.ts +32 -18
- package/cli/src/engine.ts +29 -3
- package/cli/src/generated/sotPayload.ts +14 -14
- package/cli/src/main.ts +10 -3
- package/cli/src/manifests.ts +29 -11
- package/docks-kit +6 -3
- package/docks-kit.ps1 +22 -11
- package/package.json +1 -1
package/AGENTS.md
CHANGED
|
@@ -31,7 +31,7 @@ launcher can fall back to Bun source.
|
|
|
31
31
|
| `docks-kit` / `docks-kit.ps1` | POSIX and Windows CLI launchers. On supported hosts, each runs the matching binary in `cli/dist/` only when its `--version` matches `package.json`. Otherwise it runs Bun-from-source and auto-installs Bun plus `node_modules`, and the mismatch warning names the rebuild remedy. A launcher never deletes a binary: on a host without Bun it is the only remaining recovery path. Hosts outside the support matrix fail before source fallback. The standalone platform release binary provides no-Bun recovery. |
|
|
32
32
|
| `cli/build-binaries.sh` | Compiles the six release binaries. It checksums the six known artifact names rather than globbing `docks-kit-*`, so a packed `docks-kit-<version>.tgz` is never attested as a release binary. It stamps `cli/dist/VERSION` only when the directory holds one version, and on the next build discards every artifact that stamp attributes to a different version, because a launcher refuses to run one and keeping it makes `SHA256SUMS` span two versions. An artifact with no stamp has unknown provenance and may be a hand-built or downloaded recovery binary, so the script keeps it and warns; `--prune` discards those too. Nothing here removes an artifact the stamp vouches for. |
|
|
33
33
|
| `cli/src/engine-native/` | EngineNative implementation for `sync`, `model`, and `toolchain`; idempotent, flag-gated for destructive reconciliation |
|
|
34
|
-
| `cli/src/engine-native/ompSync.ts` | omp
|
|
34
|
+
| `cli/src/engine-native/ompSync.ts` | omp sync orchestration and summary |
|
|
35
35
|
| `cli/src/commands/omp.ts` | `docks-kit omp` session launcher: renders the free-model run overlay and forwards args to omp |
|
|
36
36
|
| `cli/src/engine-native/ompOverlay.ts` | Free-model overlay render plus catalog parse and thinking-ceiling helpers |
|
|
37
37
|
| `cli/src/engine-native/<axis>.ts` | One change axis per file: `codexConfig`/`codexHooks`/`codexPlugins`/`codexStatus`, `claudeSettings`/`claudeRemovals`/`claudePluginPasses`/`claudeOptionalPlugins`/`claudeLsp`, `parseModifiers`/`parseHelp`, `ompFileDeploy`/`ompMarketplace`/`ompPlugins`, `skillsManifest`/`skillsLinks`/`skillsInstall`/`skillsPrune`, `engineCtx`/`syncConcurrency`/`syncDispatch`; orchestrators keep prior exports |
|
|
@@ -57,26 +57,28 @@ launcher can fall back to Bun source.
|
|
|
57
57
|
|
|
58
58
|
Codex SoT notes:
|
|
59
59
|
- `SoT/.codex/AGENTS.md` deploys to `~/.codex/AGENTS.md` as global Codex instructions.
|
|
60
|
-
- `SoT/.codex/config.toml` pins Codex to `model = "gpt-6.1-sol"`, sets normal and plan reasoning to `high` with concise summaries, and sets `model_verbosity = "low"`,
|
|
61
|
-
- `SoT/.codex/
|
|
60
|
+
- `SoT/.codex/config.toml` pins Codex to `model = "gpt-6.1-sol"`, sets normal and plan reasoning to `high` with concise summaries, and sets `model_verbosity = "low"`, live top-level `web_search`, workspace-write sandboxing with sandboxed command network access, cross-session `memories` (+ dedicated note tools), `[agents]` subagent limits (`max_threads = 12`, `max_depth = 2` — intentionally above Codex defaults for broad parallel kit work; deeper recursion increases cost and predictability risk), a 128 KiB `project_doc_max_bytes` budget for the repo-side AGENTS.md chain (the global `~/.codex/AGENTS.md` is uncapped and not counted), and enables the two Docks plugins `docks@docks` and `plan-lifecycle@docks` (the shared plan lifecycle).
|
|
61
|
+
- `SoT/.codex/config.toml` does not set `personality`. Codex 0.159.0 deprecated the key, and `friendly` and `pragmatic` no longer select a style. `codexConfig.ts removeRetiredTopLevelSettings, retired top-level key prune` removes the retired `personality = "pragmatic"` line from a deployed `config.toml` on every sync. A different user-set value survives.
|
|
62
|
+
- `SoT/.codex/rules/*.rules` deploys to `~/.codex/rules/` as kit-managed Codex command policy. This is Codex's equivalent of permission allow/prompt/block rules; user-learned approvals in `~/.codex/rules/default.rules` are preserved. Codex applies the strictest matching rule across all files, so a user allow rule cannot loosen a kit `forbidden` or `prompt` rule. Because the strictest rule wins, a narrow allow cannot carve an exception out of a broader prompt prefix, and a prefix rule cannot see an option after an operand. The guards therefore fail closed: every `rm`, `git rm`, `git worktree remove`, and `gio trash` prompts, and only macOS `trash` is allowed (owner decision). A `prompt` goes to `approvals_reviewer = "auto_review"`, a reviewer subagent, not always to a person. `docks.rules` carries the destructive-command guards (`docks-kit docs command-guards`); `cli/test/unit/codexSyncRules.test.ts` parses every one-line `prefix_rule`, alternatives lists and `justification` included, and asserts each guard's decision. Keep each rule on one line so that parser sees it.
|
|
62
63
|
- `SoT/.codex/plugins/marketplace.json` deploys to Codex's personal marketplace path at `~/.agents/plugins/marketplace.json`; when the `codex` CLI is available, sync reruns `codex plugin add <plugin@marketplace>` for enabled SoT plugins so stale cached installs are refreshed.
|
|
63
|
-
- Codex `/import` can copy Claude hooks into `~/.codex/hooks.json`. `
|
|
64
|
+
- Codex `/import` can copy Claude hooks into `~/.codex/hooks.json`. `codexHooks.ts removeRetiredImportedHooks, legacy SessionStart cleanup` removes only recognized hooks from retired docks-kit Claude settings, backs up a changed file, and preserves user-authored hooks. Codex `/import` (codex-rs external-agent-migration `hooks_cla.rs`) skips a handler with keys outside `type`/`command`/`timeout`/`timeoutSec`/`statusMessage`/`async`, and it skips `async: true` handlers. Thus it imports neither the kit's exec-form (`args`) SessionStart handler nor its async Notification handler.
|
|
64
65
|
- Codex limits the initial skill list to 2% of the model context and shortens descriptions first; it still reads a selected skill's full `SKILL.md`. Additive sync preserves user-added plugins, so their skill count and malformed cached `SKILL.md` files remain plugin-owned. Sync never patches a plugin cache in place.
|
|
65
66
|
- The global prompt SoTs carry the owner's standing authorization for Docks cross-company plan review, which never overrides host or platform denial.
|
|
66
67
|
- The `codex` CLI binary is upstream-owned, not kit-owned. The official standalone installer keeps package metadata under `$CODEX_HOME/packages/standalone` and places the `codex` symlink in `~/.local/bin` by default; sync only warns with a download-then-run installer command when the CLI is missing. Existing installs can self-update with `codex update`; npm and Homebrew remain upstream alternatives.
|
|
67
68
|
|
|
68
69
|
- Claude runtime settings are an authoring template with sentinels. `claudeRuntime.ts` materializes absolute Bun/script paths only after the shared `bun.ts` bootstrap is ready; `claudeSync.ts` writes all runtime assets before atomically committing settings, then prunes the legacy shell scripts and Stop hook. Native `rate_limits` is the sole quota source, so jq/curl/OAuth caches are not runtime dependencies. A missing Bun defers only this cutover and preserves legacy pointers/files.
|
|
69
70
|
- Claude's deployed SoT defaults are `model: opus` and `effortLevel: high`; `advisorModel` is deliberately absent/off. `--claude-advisor=on` is the per-machine opt-in and writes `advisorModel: opus` after the settings merge. `opus` is the alias, not a pinned id: the `minimumVersion` floor of 2.1.280 makes Claude Code resolve it to Opus 5.5, the default Opus from that release, instead of capping at Opus 5 under the former 2.1.219 floor. Keeping the alias tracks the next Opus release without a kit change and stays portable across providers. The floor also subsumes Fable 5.1's 2.1.257 requirement.
|
|
70
|
-
- Claude permission rules are pruned, not host-scoped. The kit deploys its `PowerShell(...)` deny and ask rules on every host, because Claude Code leaves the PowerShell tool opt-in off Windows rather than unavailable, and a host that enables it must already carry the guards. `claudeRetired.ts RETIRED_PERMISSION_RULES, exact retired-rule inventory` lists the rules the kit no longer ships. `
|
|
71
|
+
- Claude permission rules are pruned, not host-scoped. The kit deploys its `PowerShell(...)` deny and ask rules on every host, because Claude Code leaves the PowerShell tool opt-in off Windows rather than unavailable, and a host that enables it must already carry the guards. `claudeRetired.ts RETIRED_PERMISSION_RULES, exact retired-rule inventory` lists the rules the kit no longer ships. `claudeRemovals.ts syncRemovals, retired-permission pass` force-prunes those exact strings from the kit-managed `~/.claude/settings.json` on every sync, because `settings.ts mergeSettings, permission-array union` would otherwise keep a dropped rule forever. A different user-authored rule survives. To restore an exact retired rule for one checkout, put it in that checkout's `.claude/settings.local.json`. Claude Code resolves that file against the working directory and merges it over user settings. Sync never reads or writes the checkout-local file. Claude Code has no user-scope local settings file. For a machine-wide restoration, add the rule to `SoT/.claude/settings.json`. At the same time, remove the exact string from `claudeRetired.ts RETIRED_PERMISSION_RULES, exact retired-rule inventory`. `permissions.allow` holds four read-only entries (`Read`, `Glob`, `Grep`, and `WebSearch`) plus the working-directory edit rule `Edit(./)`. Broad shell allow rules are not restored, because an allow rule resolves before Claude Code's read-only command analyzer and the auto-mode classifier. `autoMemoryEnabled` and `autoDreamEnabled` are both `false`.
|
|
71
72
|
|
|
72
73
|
omp SoT notes:
|
|
73
74
|
- `SoT/.omp/AGENTS.md`, `config.yml`, `models.yml`, and `mcp.json` deploy to `~/.omp/agent/`.
|
|
74
|
-
- `SoT/.omp/intercom.json` deploys to `$PI_CODING_AGENT_DIR/intercom/config.json`. The default root is `~/.pi/agent`.
|
|
75
|
-
- `
|
|
75
|
+
- `SoT/.omp/intercom.json` deploys to `$PI_CODING_AGENT_DIR/intercom/config.json`. The default root is `~/.pi/agent`. Under a named omp profile (`OMP_PROFILE` or `PI_PROFILE`), it deploys to `<profile agentDir>/intercom/config.json`, because omp exports `PI_CODING_AGENT_DIR` as the profile agent directory.
|
|
76
|
+
- `ompFileDeploy.ts syncMergedYaml` deep-merges `config.yml` through `ompYaml.ts mergeOmpConfig` and `models.yml` through `mergeOmpModels`. Both wrap one generic mapping merge; only the config wrapper prunes stale `retry.fallbackChains` wildcards and merges `bash.patterns`.
|
|
77
|
+
- `SoT/.omp/config.yml` `bash.patterns` holds the destructive-command guards. The policy fails closed: only a canonical form that is safe by construction is allowed, and every other form that could force, empty, or delete asks. omp's `CRITICAL_BASH_PATTERNS` check returns a bare safety override before a matching prompt rule, and `tools.approvalMode: yolo` ignores a bare override, so a prompt on a critical command becomes an allow, in subagents too (`omp://approval-mode.md`). Every `rm` shape that check can flag is therefore a deny: recursive `rm`, `sudo` anywhere before `rm` (also by path), any `rm` with an `r` or `f` option before an absolute path, and `--no-preserve-root`. omp cannot match a literal `*`, so every other `rm` is a prompt (owner decision). The list order is deny, then three allow exceptions (`git rm -- *`, `git worktree remove -- *`, `gio trash -- *`), then prompts (including `git rm *`, `git worktree remove *`, `gio trash *`), then `trash *`. omp takes the first match, so an exception ahead of a broader prompt is the only way to exempt the `--` form, which a glob cannot exclude. `ompYaml.ts mergeBashPatterns, exception-aware rule merge` keeps user rules and groups all rules as kit deny, user deny, user prompt, kit exception (a kit allow listed before the first kit prompt), kit prompt, other user rules, kit allow, so a user deny or prompt precedes every kit allow. It treats a deployed rule equal to a kit rule (approval case-folded, `match` whitespace collapsed) as the kit copy, so a re-sync is a no-op. A rule the kit stops shipping stays as a user rule, so retire one through an explicit inventory. `cli/test/lib/ompBashPatterns.ts` ports omp 18.11.1's matcher, critical patterns, and yolo resolution; `cli/test/unit/ompBashPatterns.test.ts` asserts each command's final outcome and the merge. omp subagents reject every `prompt`, and the `eval` tool stays unrestricted by owner decision.
|
|
76
78
|
- `ompSync.ts` runs `ompRemovals.ts syncOmpRemovals, retired-key inventory` right after the config merge, and that pass force-prunes retired kit-owned keys from `~/.omp/agent/config.yml` on every sync, without `--reconcile`. The pass is required because `mergeOmpConfig` is additive, so removing a key from the SoT alone never removes it from a deployed file. A key retired with a recorded value is pruned only while the deployed value still matches that value, so a user edit survives; a key retired outright, such as `providers.webSearchOrder`, is pruned at any value.
|
|
77
79
|
- `cycleOrder` ends with `astra` as its fifth stop. `modelRoles.astra` is `openai-codex/gpt-6-astra:xhigh`, and `modelTags.astra` is visible. Astra and Fable fall back to each other through concrete selectors. `modelRoles.fable` remains `anthropic/claude-fable-5-1:medium`, with visible `modelTags.fable`. The hidden `switch_fable` role uses the same Fable selector and keeps an empty fallback chain.
|
|
78
|
-
- `modelRoles.task` is `anthropic/claude-opus-5-5:high`, and `retry.fallbackChains.task` falls back to `openai-codex/gpt-6.1-sol:high`. `task.maxEffort` is `xhigh`: a caller `effort: hi` outranks the role's `:high` suffix and maps to the top of the model ladder, so the cap keeps `hi` at `xhigh` instead of `max`. `advisor` is `anthropic/claude-opus-5-5:medium` with an empty fallback chain, because a GPT-6 Sol advisor looped on repeated reads under an Opus 5.5 session. Never put any GPT model (`openai-codex/gpt-*`, including GPT-6.1 Sol and GPT-6 Astra) in the advisor role or chain; Astra also costs too much for this role. `smol`, `commit`, and `tiny` use `openai-codex/gpt-6-luna`. The four reviewer entries in `task.agentModelOverrides`
|
|
79
|
-
- The
|
|
80
|
+
- `modelRoles.task` is `anthropic/claude-opus-5-5:high`, and `retry.fallbackChains.task` falls back to `openai-codex/gpt-6.1-sol:high`. `task.maxEffort` is `xhigh`: a caller `effort: hi` outranks the role's `:high` suffix and maps to the top of the model ladder, so the cap keeps `hi` at `xhigh` instead of `max`. `advisor` is `anthropic/claude-opus-5-5:medium` with an empty fallback chain, because a GPT-6 Sol advisor looped on repeated reads under an Opus 5.5 session. Never put any GPT model (`openai-codex/gpt-*`, including GPT-6.1 Sol and GPT-6 Astra) in the advisor role or chain; Astra also costs too much for this role. `smol`, `commit`, and `tiny` use `openai-codex/gpt-6-luna`. The kit-defined hidden `reviewer` role is `openai-codex/gpt-6.1-sol:xhigh` with `anthropic/claude-opus-5-5:xhigh` as its retry fallback, and all four reviewer entries in `task.agentModelOverrides` point to `@reviewer`. `ompOverlay.ts ROLE_ORDER, FALLBACK_ORDER` list `reviewer`, so a free session also replaces it. Only bundled `reviewer` and `security-reviewer` are discoverable OMP agents; `code-reviewer` and `plan-reviewer` stay dormant. The `SoT/toolchain.json` omp floor is 18.4.4, the oldest release the kit tested with `gpt-6.1-sol`. That release lists and serves the model through the `openai-codex` provider. `cli/docs/omp-models.md` records the checks.
|
|
81
|
+
- The seven Opus 5.5 roles (`default`, `slow`, `plan`, `designer`, `vision`, `task`, `advisor`) and the `smol`, `tiny`, `commit`, and `reviewer` retry chains use `anthropic/claude-opus-5-5` at the levels the role map records. `cli/docs/omp-models.md` carries the Artificial Analysis capture behind the role map, read 2026-09-29 at Intelligence Index v4.3.2 and Coding Agent Index v1.5 from the AA comparison-page metric tables. Opus 5.5 max has the highest index in that topic, at 58. AA has measured speed and latency for Opus 5.5 max and every GPT-6.1 Sol level; GPT-6 Luna medium is the only unmeasured speed and latency row.
|
|
80
82
|
- `modelRoles.web` is `web/firecrawl` and `retry.fallbackChains.web` carries the explicit 20-entry provider order. The kit declares both keys because the legacy `providers.webSearchOrder` key is retired: omp expands it in memory into these two keys and then drops it, and never writes that expansion back to disk. An explicit chain replaces omp's built-in web order wholesale, so every entry left out is a provider omp never tries. The owner removed the seven entries that named older models (Gemini 2.5 Flash, Claude Haiku 4.5, GPT-5.6, GPT-5.5, and Grok 4.5); keep every other provider.
|
|
81
83
|
- `SoT/.omp/models.yml` declares Astra's full `low, medium, high, xhigh, max` ladder with `defaultLevel: xhigh` as the worked provider ladder-override example. The temporary `anthropic.modelOverrides.claude-opus-5-5` block is gone, because the shared catalog now publishes the same limits, ladder, and prices. The removal switches omp from the block's `thinking.mode: effort` (a `budget_tokens` value per level from `ANTHROPIC_THINKING`, no effort value) to the catalog's `anthropic-adaptive` (adaptive thinking with the role's effort level); every kit Opus selector names its level, so the block's bare-selector `defaultLevel: high` is not needed. `ompRemovals.ts syncOmpModelRemovals, retired-block inventory` prunes it from `~/.omp/agent/models.yml` after the models merge, but only while the deployed block still equals the shipped one. `ompYaml.ts mergeOmpModels` preserves deployed-only keys in `~/.omp/agent/models.yml`, because a user file may carry provider credentials. Whole-file replacement is wrong.
|
|
82
84
|
- `SoT/.omp/AGENTS.md` carries the rule `Please remove all mannered prose.` Anthropic's Fable 5.1 prompting guide documents mannered prose as a Fable 5.1 behavior and gives that sentence as its short-version fix: https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-fable-5-1
|
|
@@ -84,9 +86,9 @@ omp SoT notes:
|
|
|
84
86
|
- `SoT/.omp/config.yml` sets `compaction.thresholdTokens` to `-1`, omp's schema default sentinel selecting reserve-based behavior: the trigger becomes `contextWindow` minus `max(floor(contextWindow * 0.15), 16384)`, which is 231,200 on the 272,000-token Codex window and 850,000 on the 1,000,000-token Anthropic window. The key ships as `-1` rather than being deleted because `ompYaml.ts mergeOmpConfig` is additive, so removing a key from SoT never removes it from a deployed file. `cli/docs/omp-context.md` (topic `omp-context`) carries the derivation and the measured evidence, so update that topic in the same commit as any compaction-setting change.
|
|
85
87
|
- Sync registers the `docks` marketplace. It installs or upgrades `docks@docks` and `plan-lifecycle@docks` at user scope.
|
|
86
88
|
- Sync installs `pi-intercom` at the verified version from `SoT/toolchain.json`.
|
|
87
|
-
- The omp CLI is upstream-owned
|
|
88
|
-
- `docks-kit omp [--model <selector>|--pick] [args...]` starts one interactive omp session
|
|
89
|
-
- The session model persists per machine in `~/.docks-kit/kit.db` (table `omp_session`), next to the harness selection. The default is `opencode-zen/muse-spark-1.3-contributor-free` at `xhigh`. `ompOverlay.ts ladderCeiling, advisorLevelFor` derive both levels from the ladder of the chosen model for the `--model` path, and `ompOverlay.ts planEffortChoice` drives the `--pick` wizard, which asks whether every role shares one level and otherwise takes one level for the main roles and one for the advisor from that model's own ladder. Levels never come from a fixed list: free ladders are not uniform, three free models publish no `medium`, and two publish no ladder, which the overlay renders as bare selectors. The picker lists only zero
|
|
89
|
+
- The omp CLI is upstream-owned. It updates through `omp update`, or through the external manager that owns the install, such as Tern or Nix. `omp update` refuses to run on an install that an external manager owns. Sync never installs or upgrades the CLI.
|
|
90
|
+
- `docks-kit omp [--model <selector>|--pick] [args...]` starts one interactive omp session. The overlay sets the 16 roles in `ompOverlay.ts ROLE_ORDER`, `judge` and `memory` included, to one free model and empties the 13 retry chains in `FALLBACK_ORDER`. Every other deployed chain stays as deployed. In particular, `modelRoles.web` and `retry.fallbackChains.web` keep the deployed web-search provider order. Thus a web search that falls through the `web/*` providers can reach the paid `openai-codex/gpt-6-luna` entry. Higher-precedence model selection can replace those overlay values. It renders the overlay to `~/.cache/docks-kit/omp-free-<model>-<digest>.yml` (mode 0600) and passes it through omp's repeatable `--config` flag. `ompOverlay.ts overlayFileName` gives each distinct model-and-level combination its own file, and the digest covers the selector and both levels. omp can re-read the overlay during a live session, so a second launcher with another model or level must not rewrite it. Deployed `~/.omp/agent/` files stay untouched, so the next plain `omp` run uses the paid configuration again. Remaining arguments forward verbatim to omp.
|
|
91
|
+
- The session model persists per machine in `~/.docks-kit/kit.db` (table `omp_session`), next to the harness selection. The default is `opencode-zen/muse-spark-1.3-contributor-free` at `xhigh`. Every launch, a plain launch included, runs `omp models --json` (as `omp --profile <name> models --json` when the forwarded arguments select a profile) and checks the remembered or default model against the live free catalog. When that model is not in the catalog, the launcher stops with an error and does not start omp. `ompOverlay.ts ladderCeiling, advisorLevelFor` derive both levels from the ladder of the chosen model for the `--model` path, and `ompOverlay.ts planEffortChoice` drives the `--pick` wizard, which asks whether every role shares one level and otherwise takes one level for the main roles and one for the advisor from that model's own ladder. Levels never come from a fixed list: free ladders are not uniform, three free models publish no `medium`, and two publish no ladder, which the overlay renders as bare selectors. The picker lists only catalog models at zero input and output cost with `pricingStatus` free or unknown.
|
|
90
92
|
- `docks-kit omp --fast` toggles `tier.openai` in the deployed `~/.omp/agent/config.yml` between `priority` (omp's `/fast on` tier for OpenAI and OpenAI-Codex models) and `none`, prints the change, and exits. `ompFastMode.ts toggleOpenAiFast` edits only that key. `SoT/.omp/config.yml` does not declare `tier.openai`, so the additive merge keeps the toggled value across sync. There is no Ultrafast flag: omp ignores `ultrafast` for a Codex model whose catalog `serviceTiers` omits it, and the current GPT models omit it.
|
|
91
93
|
|
|
92
94
|
For per-tool SoT layouts (`SoT/.claude/`, `SoT/.codex/`, `SoT/.omp/`), see the matching SoT directory.
|
|
@@ -94,14 +96,14 @@ For per-tool SoT layouts (`SoT/.claude/`, `SoT/.codex/`, `SoT/.omp/`), see the m
|
|
|
94
96
|
## Engineering rules
|
|
95
97
|
|
|
96
98
|
- **Idempotent operations.** Every EngineNative sync step must be safe to re-run. Settings merges, plugin installs, and marketplace adds are all idempotent — re-running with no SoT changes is a no-op.
|
|
97
|
-
- **Failed harness-CLI operations fail the sync.** A kit-managed marketplace or plugin command that exits non-zero is recorded through `failures.ts recordFailure` — it still warns in place, and `
|
|
99
|
+
- **Failed harness-CLI operations fail the sync.** A kit-managed marketplace or plugin command that exits non-zero is recorded through `failures.ts recordFailure` — it still warns in place, and `syncDispatch.ts engineSync, failure ledger` lists every entry under `--- Failures ---` after the summary, then returns exit 1. A skip is not a failure: a missing harness CLI, missing git, and a missing `SoT/toolchain.json` pin stay warn-only with exit 0; an unavailable plugin inventory warns without a ledger row and falls back to the full refresh, whose failed commands are recorded. `ompMarketplace.ts syncMarketplace, probe guard` returns before spawning so a host without omp never records one. Aggregate roll-up warnings (`N plugin operation(s) failed`) stay warns, because each counted site already records itself.
|
|
98
100
|
- **Removed bash engine.** The bash engine was removed after the `bash-engine-final` tag. `DOCKS_KIT_ENGINE=bash` must fail with the removed-engine message; engine bugs are fixed forward in EngineNative.
|
|
99
101
|
- **Effect 4 CLI stack.** The CLI pins `effect@4.0.0-rc.115` (including `effect/unstable/cli`), `@effect/platform-bun@4.0.0-rc.115` (`BunServices.layer`, `BunRuntime.runMain`), `@effect/vitest@4.0.0-rc.115`, and `vitest@5.0.0` (inside the `@effect/vitest` peer range `>=5.0.0 <6.0.0`; 5.0.0 is past the 5.x patch point for GHSA-82fw-gwwq-j7x9, fixed in 5.0.0-rc.2). `@effect/cli` and `@effect/platform` are removed and must not be reintroduced.
|
|
100
102
|
- **Effect skill routing.** Effect work in this checkout must verify migration and API call shapes against the installed declarations under `node_modules/effect/dist/unstable/cli/`, never from memory or a mutable dist-tag. The `effect-ts-setup`, `effect-ts-port`, and `effect-ts-specialist` skills target Effect 3.x and do not apply.
|
|
101
103
|
- **Targeted syncs.** `./docks-kit sync` accepts positional targets: `claude`, `codex`, `agents`, and `omp`. Use the narrowest target that matches the SoT change (for example, `./docks-kit sync omp` for omp-only config edits); targets can be combined with `--dry-run`, `--skip-bubblewrap` (skip optional bubblewrap bootstrap for the Codex Linux sandbox), `--skip-plugin-refresh` (install missing plugins without refreshing existing caches), `--reconcile`, `--prune`, and the deploy-time modifiers `--claude-compact-window=<tokens>` / `--claude-permissive` / `--claude-model=<m>` / `--claude-effort=<level>` / `--claude-advisor=<on|off|default>` / `--codex-model=<m>` / `--codex-effort=<level>` (see `CLAUDE.md` § Deploy-time modifiers).
|
|
102
104
|
- **Per-machine harness selection.** The SQLite store `~/.docks-kit/kit.db` drives a flag-less sync. A missing selection selects `claude`, `codex`, and `agents`; it never selects `omp` implicitly. `sync` never prompts and never writes the selection. `docks-kit harnesses` is the only command that writes the `harness_selection` table. With no stored selection, its picker pre-ticks all four harnesses, `omp` included, so a new machine opts out of omp instead of opting in; nothing changes until the picker saves. The same store holds the independent `omp_session` table, which only `docks-kit omp` writes, and a TTL cache (`cache_entry`) for the live Anthropic model list and `toolchain outdated` lookups. `kitDb.ts withKitDb` is the only opener: it runs the append-only `MIGRATIONS`, imports a legacy `state.json` once and renames it to `state.json.migrated`, and refuses a store from a newer schema. Cache rows carry the kit version as fingerprint, so a new release refetches them.
|
|
103
|
-
- **Additive by default.** Keys present in deployed config but absent from SoT are preserved on default sync. This protects user-only additions, but means drift accumulates — neither flag-less reset can clean it up. The one exception is the Claude `removed` manifest (`
|
|
104
|
-
- **`--reconcile` / `--prune` are the kit-owned reconcile flags.** Orthogonal — `--reconcile` reconciles the settings layer (SoT-declared keys/tables/arrays win; user-only keys and nested objects are preserved; permissions arrays are replaced wholesale by SoT). `--prune`
|
|
105
|
+
- **Additive by default.** Keys present in deployed config but absent from SoT are preserved on default sync. This protects user-only additions, but means drift accumulates — neither flag-less reset can clean it up. The one exception is the Claude `removed` manifest (`claudeRemovals.ts REMOVED_MANIFEST, baseline removal inventory`), a curated list of unambiguous kit-owned artifacts that `claudeRemovals.ts syncRemovals, baseline artifact prune` force-prunes on every sync, including the home-relative `~/.local/bin/session-relay` artifact installed outside `~/.claude`; see `CLAUDE.md` § Pruning stale artifacts. Codex `config.toml` merges at table granularity: a table the SoT declares (`[features]`, `[sandbox_workspace_write]`, `[tui]`, ...) is replaced wholesale, so user keys inside it are dropped. The prior file stays in `config.toml.bak`, and user-only top-level keys and tables survive.
|
|
106
|
+
- **`--reconcile` / `--prune` are the kit-owned reconcile flags.** Orthogonal — `--reconcile` reconciles the settings layer (SoT-declared keys/tables/arrays win; user-only keys and nested objects are preserved; permissions arrays are replaced wholesale by SoT). `--prune` removes what SoT does not declare: every user-scope Claude plugin whose id is absent from SoT `enabledPlugins` (third-party and previously opted-in plugins included, unless this run passes their `--claude-plugin` flag), every Claude marketplace except `claude-plugins-official`, SoT/opt-in plugin marketplaces, and marketplaces referenced by project/local installs, and the `~/.agents/skills/*` entries tracked in `~/.agents/.kit-managed-skills` that SoT no longer lists. Combine with `--reconcile` for a full reset to SoT. Always preserved: user-only settings keys (custom env vars, mcpServers), manually-installed skills, and project/local-scope plugin installs. Each tool's per-tool file documents the specific paths and diff recipes.
|
|
105
107
|
- **SOLID-aligned modules.** Axis modules own one concern each (see layout); orchestrators (`claudeSync.ts`, `codexSync.ts`, `skillsSync.ts`, `ompSync.ts`, `index.ts`, `argv.ts`, `parseArgs.ts`) keep prior exports. New shared shapes go in `sharedTypes.d.ts`, never a second local copy. The public CLI seam is `cli/src/engine.ts`.
|
|
106
108
|
- **Small, reviewable changes.** Bundled multi-concern PRs are harder to review and revert. Split an engine/CLI change and a per-tool config change unless the change requires atomicity.
|
|
107
109
|
- **Version bumps.** Bump `package.json` on user-visible behavior change; bump a `SoT/toolchain.json` `verified` pin only after testing that release. Refactors, splits, and doc edits bump nothing.
|
|
@@ -120,7 +122,7 @@ For per-tool SoT layouts (`SoT/.claude/`, `SoT/.codex/`, `SoT/.omp/`), see the m
|
|
|
120
122
|
- No secrets in SoT. The kit's SoT directories are committed; treat them as declarative config only.
|
|
121
123
|
- Treat external plugin marketplaces and installer downloads as untrusted input. Prefer download-then-run over `curl … | bash` so a truncated stream cannot execute a partial script.
|
|
122
124
|
- **Pin, never float.** Every kit-driven install of third-party software is pinned to a `SoT/toolchain.json` `verified` version — no `@latest` npm/bun installs (Shai-Hulud-class worm surface), no mutable action tags in workflows (commit SHAs only), installer scripts fetched from version tags where upstream supports it. New third-party install surface ⇒ manifest pin first. The rule also covers a transitive dependency whose declared range is a caret or tilde over a prerelease: `^4.0.0-rc.109` admits every later prerelease of the same version tuple, and a published install resolves that range fresh because `bun.lock` is not packaged, so a runtime pinned in `package.json` meets a newer sibling and the process fails at startup. The fix is an exact root pin plus both guards: the offline `cli/test/unit/dependencyPins.test.ts` and the clean-room `bun run smoke:package`. Moving the Effect release is therefore one lockstep commit: set `effect`, `@effect/platform-bun`, `@effect/platform-node-shared`, and `@effect/vitest` to the same version together, then run `bun run smoke:package`; the guard reports `unsatisfied` while one of them lags. Details: the `toolchain-context` skill and `cli/docs/toolchain.md`.
|
|
123
|
-
- **One exemption: the kit's own package.** `install.sh` and `install.ps1` end with `bun add -g docks-kit@latest`, because a global installer that pinned itself would install a fixed old kit forever, and pinning it to `package.json` would request an unpublished version between the release-prep commit and the npm publish. The exemption covers `docks-kit` alone. Both installers still pin the Bun installer they download to the manifest's verified version
|
|
125
|
+
- **One exemption: the kit's own package.** `install.sh` and `install.ps1` end with `bun add -g docks-kit@latest`, because a global installer that pinned itself would install a fixed old kit forever, and pinning it to `package.json` would request an unpublished version between the release-prep commit and the npm publish. The exemption covers `docks-kit` alone. Both installers still pin the Bun installer they download to the manifest's verified version. `cli/test/unit/hostTargets.test.ts` asserts that pin in all four launcher and installer scripts, and `cli/test/unit/install.test.ts` proves `install.sh` passes it to the Bun installer.
|
|
124
126
|
## Testing
|
|
125
127
|
|
|
126
128
|
Run `bun run check` before and after each change: oxlint, `oxfmt --check`, `tsc --noEmit`, unit tests, and both golden suites. Scripts: `lint`, `format:check`, `typecheck`, `test:unit`, `test:integration` (goldens), `test`, `format`; CI runs `check` on push and pull request with Bun 1.4.2. Prove-red modes must exit non-zero after detecting planted mismatches. The golden writer emits two-space JSON that `oxfmt` re-serializes, so run `bun run format` after `--update-goldens` or `format:check` fails. Also verify user-facing changes via `./docks-kit sync --dry-run`, per-tool sanity (`/doctor`, `/plugin`, etc.), `bun run test:runtime:posix`, and `diff <(jq -S . <SoT>) <(jq -S . <deployed>)` recipes from the per-tool file.
|
|
@@ -159,7 +161,7 @@ Use `unslop` only for its pattern-detection lists.
|
|
|
159
161
|
When a kit-mechanic skill, its `references/`, or a wrapper agent (`.claude/agents/*.md` + its `.codex/agents/*.toml` twin) cites EngineNative internals, name the **module + exported/local function + semantic anchor** (e.g. `claudePluginPasses.ts syncPlugins, pass 5 uninstall guard`) — never a raw `file:NNN` line number, which goes stale on every refactor. Keep exactly one coarse `metadata.source_files[].lines` range per skill file as the sole intentional line-number touchpoint.
|
|
160
162
|
</constraint>
|
|
161
163
|
|
|
162
|
-
**Universal-skill bootstrap.** `SoT/.agents/skills.txt` is intentionally empty, so the default sync exposes no universal skills. The generic opt-in contract remains: each [agentskills.io](https://agentskills.io/specification) slug is installed by `
|
|
164
|
+
**Universal-skill bootstrap.** `SoT/.agents/skills.txt` is intentionally empty, so the default sync exposes no universal skills. The generic opt-in contract remains: each [agentskills.io](https://agentskills.io/specification) slug is installed by `skillsInstall.ts syncUniversal` with `npx --yes skills@<verified> add <slug> -g -y -a claude-code codex`, where `<verified>` is the `skills-cli` pin in `SoT/toolchain.json`. The slug comes first because `-a/--agent` is variadic; naming both supported agents preserves the canonical `~/.agents/skills/<name>/SKILL.md` plus Claude's `~/.claude/skills/<name>` symlink. Existing canonical directories are reused idempotently, `--prune` removes only entries tracked in `~/.agents/.kit-managed-skills`, and skills requiring a separate CLI retain explicit helpers in `skillsSync.ts`.
|
|
163
165
|
|
|
164
166
|
## Plans
|
|
165
167
|
|
package/README.md
CHANGED
|
@@ -36,9 +36,10 @@ The executable carries the generated sync payload; no checkout or adjacent
|
|
|
36
36
|
`SoT/` directory is required.
|
|
37
37
|
|
|
38
38
|
Source and global installs require Bun. npm-global tools require Node/npm.
|
|
39
|
-
`jq` is optional doctor/test tooling. `curl`
|
|
40
|
-
|
|
41
|
-
|
|
39
|
+
`jq` is optional doctor/test tooling. `curl` is needed only to download Bun's
|
|
40
|
+
upstream installer when Bun is absent: by the POSIX launcher and installer, and
|
|
41
|
+
by sync's Bun bootstrap, which a release binary can also run. The Windows
|
|
42
|
+
launcher and installer use Invoke-WebRequest.
|
|
42
43
|
|
|
43
44
|
## CLI
|
|
44
45
|
|
|
@@ -54,7 +55,7 @@ docks-kit toolchain [check|ensure <tool>|outdated|upgrade] verified-version flo
|
|
|
54
55
|
docks-kit status [--json] deployed-vs-SoT drift + toolchain + counts
|
|
55
56
|
docks-kit plugins list [--json] enabledPlugins tri-state vs installed
|
|
56
57
|
docks-kit skills list [--json] universal skills vs manifest
|
|
57
|
-
docks-kit docs [topic] self-documentation (
|
|
58
|
+
docks-kit docs [topic] self-documentation (12 topics)
|
|
58
59
|
--help --version --wizard --completions built-in
|
|
59
60
|
```
|
|
60
61
|
|
|
@@ -109,6 +110,16 @@ and a later flag-less sync reverts them. Full reference: `docks-kit docs flags`
|
|
|
109
110
|
Claude's native `rate_limits`; there is no OAuth fetch, shared usage cache,
|
|
110
111
|
or Stop hook. If Bun cannot be resolved or bootstrapped, sync preserves a
|
|
111
112
|
working legacy hook/statusline setup and reports that migration is deferred.
|
|
113
|
+
- **Command guards** — omp, Claude Code, and Codex share one fail-closed
|
|
114
|
+
policy for destructive shell commands; to remove, agents use the trash
|
|
115
|
+
(`gio trash -- <path>`, or `trash <path>` on macOS). `rm` of the home
|
|
116
|
+
directory or `/` and `HOME` reassignment are denied. omp also denies every
|
|
117
|
+
recursive `rm` and asks before any other `rm`. `find -delete`, `git clean`,
|
|
118
|
+
`git reset --hard`, any `git push` that could force, and `git rm` other
|
|
119
|
+
than `git rm -- <path>` ask; Claude Code and Codex also ask before a
|
|
120
|
+
recursive `rm` and every `git rm`. An omp subagent cannot answer an ask, so
|
|
121
|
+
omp rejects it; a Codex ask can go to its `auto_review` reviewer. Details,
|
|
122
|
+
per-tool differences, and gaps: `docks-kit docs command-guards`.
|
|
112
123
|
|
|
113
124
|
## Repository map
|
|
114
125
|
|
|
@@ -155,8 +166,8 @@ are versioned config snapshots without shipping the authoring `SoT/` tree.
|
|
|
155
166
|
## Deeper docs
|
|
156
167
|
|
|
157
168
|
- `docks-kit docs <topic>` - overview, sync-layers, flags, modifiers, models,
|
|
158
|
-
omp-models, toolchain, plugins, install,
|
|
159
|
-
with the CLI)
|
|
169
|
+
omp-models, omp-context, command-guards, toolchain, plugins, install,
|
|
170
|
+
platforms (works offline, bundled with the CLI)
|
|
160
171
|
- [`AGENTS.md`](AGENTS.md) — engineering rules for agents working on the kit
|
|
161
172
|
- [`CLAUDE.md`](CLAUDE.md) — Claude Code specifics: env vars, session
|
|
162
173
|
management, permission mode, open concerns
|
|
@@ -0,0 +1,269 @@
|
|
|
1
|
+
# Command guards
|
|
2
|
+
|
|
3
|
+
The kit ships one destructive-command policy to omp, Claude Code, and Codex.
|
|
4
|
+
The short rule for agents: **to remove, use the trash**. Run
|
|
5
|
+
`gio trash -- <path>` on Linux or `trash <path>` on macOS. To remove a path
|
|
6
|
+
from the git index, run `git rm -- <path>`.
|
|
7
|
+
|
|
8
|
+
The policy follows one principle: **fail closed**. A command is allowed only
|
|
9
|
+
in a canonical form that is safe by construction. Every other form that could
|
|
10
|
+
force, empty, or delete asks. Where omp's critical-command check acts, the
|
|
11
|
+
form is denied instead. A false positive in the ask tier is accepted.
|
|
12
|
+
|
|
13
|
+
The policy sorts shell commands into three tiers:
|
|
14
|
+
|
|
15
|
+
- **Deny**: the command never runs, and nobody can approve it in the session.
|
|
16
|
+
- **Ask**: the command needs approval before it runs.
|
|
17
|
+
- omp asks a person in the main session. omp runs subagents headless, so it
|
|
18
|
+
rejects every ask in a subagent.
|
|
19
|
+
- Claude Code asks a person, even in auto mode.
|
|
20
|
+
- Codex sends the prompt to the approvals reviewer in the kit's
|
|
21
|
+
`config.toml`. That reviewer is `auto_review`, a reviewer subagent, so a
|
|
22
|
+
Codex ask can be approved without a person.
|
|
23
|
+
- **Allow**: the agent may run the command alone.
|
|
24
|
+
|
|
25
|
+
The omp `eval` tool stays unrestricted: `bash.patterns` does not gate a shell
|
|
26
|
+
that `eval` starts.
|
|
27
|
+
|
|
28
|
+
## Tiers per tool
|
|
29
|
+
|
|
30
|
+
| Command | omp `bash.patterns` | Claude Code `permissions` | Codex `docks.rules` |
|
|
31
|
+
| --- | --- | --- | --- |
|
|
32
|
+
| `rm` of `/`, `~`, `~/`, `$HOME`, `${HOME}`, quoted or with a trailing `/` | deny | deny, also through `/bin/rm` and other `*/rm` paths | deny `/` as argv; deny the exact scripts listed in `docks.rules` |
|
|
33
|
+
| `rm` of `/*` | deny with an `r` or `f` option; plain `rm /*` asks | deny | deny the exact scripts listed |
|
|
34
|
+
| `HOME=...`, `export HOME=...`, `env HOME=...` | deny | deny `export`, `env`, and a bare `HOME=...` line; `HOME=... cmd` goes to the classifier | deny `export HOME=$(mktemp -d)` and three spellings; `env` asks |
|
|
35
|
+
| `sudo` before `rm`, plain or by path (`/usr/bin/sudo`) | deny | deny (`sudo *`, `*/sudo *`) | deny (`sudo`, resolved by path) |
|
|
36
|
+
| `git push --force` or `-f` to `origin main` or `origin master` | deny | deny | deny |
|
|
37
|
+
| Recursive `rm` (`-r`, `-R`, `-rf`, `-fr`, `-rfv`, `--recursive`, other listed clusters) | deny | ask | ask (every `rm` asks) |
|
|
38
|
+
| `rm` with an `r` or `f` option before an absolute path, `--no-preserve-root` | deny | ask if recursive, else classifier | ask |
|
|
39
|
+
| Any other `rm`: one file, several files, a glob, `-f` | ask | glob asks; the rest goes to the classifier | ask |
|
|
40
|
+
| `find ... -delete`, `find ... -exec rm` | ask | ask | ask (every `find` asks) |
|
|
41
|
+
| `git clean`, `git reset --hard` | ask | ask | ask |
|
|
42
|
+
| `git push` that could force: see below | ask | ask | ask (every `git push` asks) |
|
|
43
|
+
| `git rm -- <path>`, `git worktree remove -- <path>` | allow | ask | ask |
|
|
44
|
+
| Any other `git rm` or `git worktree remove` | ask | ask | ask |
|
|
45
|
+
| `gio trash -- <path>` | allow | allow | ask |
|
|
46
|
+
| Any other `gio trash` | ask | ask with `--empty` or `--restore`; the rest goes to the classifier | ask |
|
|
47
|
+
| `trash <path>` (macOS) | allow | allow | allow |
|
|
48
|
+
|
|
49
|
+
A `git push` asks when it holds any of these words:
|
|
50
|
+
|
|
51
|
+
- a `+` refspec;
|
|
52
|
+
- any long option that starts with `--f`, such as `--force`,
|
|
53
|
+
`--force-with-lease`, or an abbreviation like `--for`;
|
|
54
|
+
- `--mirror` or an abbreviation of it;
|
|
55
|
+
- `-f`, or a short-option cluster that starts with two of git push's
|
|
56
|
+
no-argument flags (`v`, `q`, `d`, `n`, `u`, `4`, `6`, `f`), such as `-vf`,
|
|
57
|
+
`-vvf`, or `-uvfq`;
|
|
58
|
+
- a word that starts with `$` or a backtick. In Claude Code, also a word that
|
|
59
|
+
starts with a quote or a backslash.
|
|
60
|
+
|
|
61
|
+
Each of these must begin a word: `feat/cache-vf` is a branch name, not an
|
|
62
|
+
option. `--follow-tags` also asks. This is an accepted false positive, because
|
|
63
|
+
it shares the `--f` prefix.
|
|
64
|
+
|
|
65
|
+
## Why the tools differ
|
|
66
|
+
|
|
67
|
+
| Topic | omp | Claude Code | Codex |
|
|
68
|
+
| --- | --- | --- | --- |
|
|
69
|
+
| Rule order | First match wins | Deny, then ask, then allow, whatever the order | Strictest matching rule wins |
|
|
70
|
+
| Exception for one form | An allow before a broader prompt works, so `git rm -- *` is allowed before `git rm *` asks | An allow never beats a matching ask, so every `git rm` asks | An allow never beats a matching prompt, so every `gio trash` asks |
|
|
71
|
+
| Pattern | Glob on the whole line and on each quote-stripped segment; `*` crosses spaces; no literal `*` | Glob on each subcommand as written, quotes included; `\*` is a literal `*` | Exact argv tokens from the start, with alternatives |
|
|
72
|
+
| `rm` by path (`/bin/rm`) | Matched by `*/rm` forms | Matched by `*/rm` forms | Resolved to the `rm` rules |
|
|
73
|
+
| Leading assignment `X=1 cmd` | Part of the line | Stripped before deny and ask rules | A script with it is not split |
|
|
74
|
+
| What an ask means | A person in the main session; rejected in a subagent | A person | The `auto_review` reviewer subagent |
|
|
75
|
+
| Critical-command check | Turns a prompt into an allow under yolo, so the kit denies those shapes | Asks or denies for an `rm` of a critical path | Prompts for a forced `rm` that no rule matches |
|
|
76
|
+
|
|
77
|
+
## How an agent removes files
|
|
78
|
+
|
|
79
|
+
1. Move a file or a directory to the trash with `gio trash -- <path>` on Linux
|
|
80
|
+
or `trash <path>` on macOS. The `--` ends option parsing, so `gio` reads
|
|
81
|
+
every later argument as a path, and a path such as `--empty` cannot empty
|
|
82
|
+
the trash.
|
|
83
|
+
2. `gio trash` does not trash on a system internal mount, such as a `tmpfs`
|
|
84
|
+
`/tmp`. Leave a temporary file in `/tmp`; the system clears it.
|
|
85
|
+
3. To remove a path from the git index, run `git rm -- <path>`. To remove a
|
|
86
|
+
worktree, run `git worktree remove -- <path>`. Never pass `--force`.
|
|
87
|
+
4. In omp, do not use `rm`. omp denies a recursive `rm` and asks before any
|
|
88
|
+
other `rm`, so a subagent cannot run `rm` at all.
|
|
89
|
+
5. In Claude Code and Codex, use `rm <file>` without flags only for one file
|
|
90
|
+
that the agent created.
|
|
91
|
+
6. For a recursive or bulk deletion, ask the user in the main session. A
|
|
92
|
+
subagent lists the paths in its final report. In omp, the user runs a
|
|
93
|
+
recursive deletion by hand.
|
|
94
|
+
7. Write literal absolute paths. Never use `~` or `$HOME`, and never set or
|
|
95
|
+
export `HOME`.
|
|
96
|
+
|
|
97
|
+
The deployed agent prompts (`~/.omp/agent/AGENTS.md`, `~/.claude/CLAUDE.md`,
|
|
98
|
+
and the Codex `AGENTS.md`) carry the same rules.
|
|
99
|
+
|
|
100
|
+
### What the git removals keep
|
|
101
|
+
|
|
102
|
+
- `git rm` without `--force` refuses a file whose content differs from the
|
|
103
|
+
last commit. The removed content stays in the repository history.
|
|
104
|
+
- `git worktree remove` without `--force` refuses a worktree with untracked
|
|
105
|
+
files or tracked changes. It does delete the files that `.gitignore` covers,
|
|
106
|
+
such as `node_modules`, build output, or a local `.env`. Git keeps no copy of
|
|
107
|
+
those files.
|
|
108
|
+
- Git accepts an abbreviated long option, such as `--for` for `--force`, and a
|
|
109
|
+
short-option cluster, such as `-rff`. After `--`, git reads every argument as
|
|
110
|
+
a path. So the `--` form is the only one that cannot force.
|
|
111
|
+
|
|
112
|
+
## Why omp denies instead of asking
|
|
113
|
+
|
|
114
|
+
omp has a built-in check for critical commands. For example, it flags:
|
|
115
|
+
|
|
116
|
+
- an `rm` with an `r` or `f` option before an absolute path;
|
|
117
|
+
- `sudo rm`, also when `sudo` is called by its path.
|
|
118
|
+
|
|
119
|
+
For a critical command, omp returns a bare safety override before it returns
|
|
120
|
+
a matching prompt rule. Under the kit's `tools.approvalMode: yolo`, omp
|
|
121
|
+
ignores a bare override, so the command runs. This applies in subagents too.
|
|
122
|
+
Only a deny rule stops a critical command (`omp://approval-mode.md`).
|
|
123
|
+
|
|
124
|
+
The kit therefore denies every `rm` shape that the check can flag:
|
|
125
|
+
|
|
126
|
+
- a recursive `rm`;
|
|
127
|
+
- an `rm` with an option that holds `r` or `f` before an absolute path;
|
|
128
|
+
- `sudo` anywhere before `rm`;
|
|
129
|
+
- `--no-preserve-root`.
|
|
130
|
+
|
|
131
|
+
## Matching rules per tool
|
|
132
|
+
|
|
133
|
+
- **omp**:
|
|
134
|
+
- The rules are ordered, and the first match wins.
|
|
135
|
+
- `*` matches any text, spaces included, and no escape spells a literal `*`.
|
|
136
|
+
- Deny and prompt rules test the whole line and each segment between `;`,
|
|
137
|
+
`&&`, `|`, and similar operators, with quotes and escapes removed. So
|
|
138
|
+
`rm -rf "$HOME"` matches `rm * $HOME`.
|
|
139
|
+
- Allow rules apply only to a simple command.
|
|
140
|
+
- The kit lists rules as deny, then the three `-- *` allow exceptions, then
|
|
141
|
+
prompts, then the `trash *` allow.
|
|
142
|
+
- **Claude Code**:
|
|
143
|
+
- Deny beats ask, and ask beats allow, whatever the order.
|
|
144
|
+
- The rules match the command text as written, quotes included, so the kit
|
|
145
|
+
lists each quoted spelling. `\*` is a literal `*`.
|
|
146
|
+
- Claude Code strips wrappers such as `timeout` and bare `xargs`. It matches
|
|
147
|
+
deny and ask rules past a leading variable assignment, and it applies them
|
|
148
|
+
to every subcommand.
|
|
149
|
+
- A rule for `rm` does not match `/bin/rm`, so the kit adds `*/rm` forms.
|
|
150
|
+
- **Codex**:
|
|
151
|
+
- A rule matches argv tokens from the start, and the strictest matching rule
|
|
152
|
+
wins.
|
|
153
|
+
- Codex splits a `bash -lc` script into argv only when every word is a plain
|
|
154
|
+
literal. A script with `~`, `$HOME`, a glob, or a substitution stays one
|
|
155
|
+
`[shell, -lc, script]` invocation, so a rule can match only that exact
|
|
156
|
+
script.
|
|
157
|
+
- Codex resolves `/bin/rm` and `/usr/bin/sudo` to the `rm` and `sudo` rules.
|
|
158
|
+
|
|
159
|
+
## Gaps
|
|
160
|
+
|
|
161
|
+
- **omp, literal `*`**: omp cannot match a literal `*`, so it cannot tell a
|
|
162
|
+
glob from a file name. The kit therefore asks before every `rm` that it does
|
|
163
|
+
not deny.
|
|
164
|
+
- **omp, whole-line matching**: omp tests deny rules against the whole command
|
|
165
|
+
line too, so a pattern can span commands. For example, `rm x && cd ~`,
|
|
166
|
+
`rm -i a && ls -R /tmp`, and `sudo apt update && rm x` are denied. Run `rm`
|
|
167
|
+
as its own command.
|
|
168
|
+
- **omp, text arguments**: the critical-shape deny also matches text such as
|
|
169
|
+
`git commit -m "rm -f /tmp/x"`. Write such a message with `git commit -F`.
|
|
170
|
+
- **omp, critical patterns in a compound line**: a line that holds one of
|
|
171
|
+
omp's other critical patterns, such as `curl ... | sh`, turns any matching
|
|
172
|
+
prompt rule on that line into an allow.
|
|
173
|
+
- **omp, wrapped `rm`**: a recursive `rm` behind `xargs`, `env`, `nice`, or
|
|
174
|
+
`timeout` asks instead of being denied. With an option before an absolute
|
|
175
|
+
path, or behind `sudo`, it is denied.
|
|
176
|
+
- **omp, user prompts and the exceptions**: user deny and prompt rules come
|
|
177
|
+
before the kit's `-- *` allow exceptions, so a user rule can still tighten
|
|
178
|
+
them.
|
|
179
|
+
- **Recursive `rm` clusters**: omp and Claude Code match only the listed
|
|
180
|
+
recursive-`rm` clusters: `r` or `R` alone or with `f` and `v`, in any order.
|
|
181
|
+
Another cluster, such as `rm -ri`, falls back to the tool's next tier. In
|
|
182
|
+
omp, that tier is ask. In Claude Code, the classifier decides.
|
|
183
|
+
- **`git push` beyond the listed words**: a configured `+` refspec in
|
|
184
|
+
`remote.<name>.push` makes a plain `git push` force. No rule sees git
|
|
185
|
+
configuration. `git push --delete` and `git push --prune` remove remote
|
|
186
|
+
branches without force; they ask only in Codex.
|
|
187
|
+
- **`git -C`**: omp and Claude Code do not match a git subcommand that does
|
|
188
|
+
not follow `git` directly, such as `git -C . push --force`.
|
|
189
|
+
- **Claude Code, `HOME=` before a command**: Claude Code matches deny rules
|
|
190
|
+
past a leading assignment. So `HOME=/tmp/x cmd` reaches the classifier.
|
|
191
|
+
`export HOME=...`, `env HOME=...`, a bare `HOME=...` line, and any `rm` of
|
|
192
|
+
`$HOME` stay denied.
|
|
193
|
+
- **Claude Code, unlisted `rm` forms**: a multi-file `rm` and an `rm -f` of
|
|
194
|
+
one file go to the auto-mode classifier. Claude Code does not strip `env`, so
|
|
195
|
+
`env X=1 rm -rf ~` also goes to the classifier.
|
|
196
|
+
- **Claude Code, `gio trash`**: a `gio trash` form other than `-- <path>` that
|
|
197
|
+
holds neither `--empty` nor `--restore` goes to the classifier. GLib's
|
|
198
|
+
`gio trash` supports no abbreviated long options.
|
|
199
|
+
- **Codex, scripts**: a home or root deletion inside a longer script, such as
|
|
200
|
+
`cd x && rm -rf ~`, falls to Codex's built-in forced-`rm` prompt and to the
|
|
201
|
+
sandbox. Codex matches `HOME=` reassignment only as an exact script, and
|
|
202
|
+
`export HOME=/tmp` runs sandboxed. Each Codex command runs in a new shell.
|
|
203
|
+
- **Codex, prompt reviewer**: a Codex prompt reaches the `auto_review`
|
|
204
|
+
reviewer subagent, not always a person.
|
|
205
|
+
- **All tools**: no rule covers a program that removes files itself, such as a
|
|
206
|
+
Python or Node script, or a shell started by omp `eval`.
|
|
207
|
+
|
|
208
|
+
### How `gio trash` empties the trash
|
|
209
|
+
|
|
210
|
+
GLib's `gio trash` handles operands first. With at least one path, it trashes
|
|
211
|
+
or restores the paths, and `--empty` has no effect. Without a path, it empties
|
|
212
|
+
the trash when `--empty` is set. `-f` can repeat. So `gio trash -ff --empty`
|
|
213
|
+
empties the trash, while `gio trash build --empty` only trashes `build`.
|
|
214
|
+
|
|
215
|
+
In Codex, every `gio trash` asks, because a prefix rule cannot tell
|
|
216
|
+
`-ff --empty` from `-- <path>`. In omp, every form but `gio trash -- <path>`
|
|
217
|
+
asks. In Claude Code, any form with `--empty` or `--restore` asks.
|
|
218
|
+
|
|
219
|
+
## Sync behavior
|
|
220
|
+
|
|
221
|
+
- **omp**:
|
|
222
|
+
- The config merge keeps user `bash.patterns` rules and groups all rules in
|
|
223
|
+
this order:
|
|
224
|
+
1. kit deny;
|
|
225
|
+
2. user deny;
|
|
226
|
+
3. user prompt;
|
|
227
|
+
4. kit allow exceptions (the kit allows listed before the first kit
|
|
228
|
+
prompt);
|
|
229
|
+
5. kit prompt;
|
|
230
|
+
6. the other user rules;
|
|
231
|
+
7. kit allow.
|
|
232
|
+
- A user deny or prompt thus precedes every kit allow, and no allow can
|
|
233
|
+
loosen a deny or a prompt.
|
|
234
|
+
- A repeat sync changes nothing.
|
|
235
|
+
- **Claude Code**: sync unions each permissions array with the deployed one,
|
|
236
|
+
so user rules stay. `--reconcile` replaces the arrays with the SoT arrays.
|
|
237
|
+
- **Codex**: sync writes `~/.codex/rules/docks.rules` whole. Keep your own
|
|
238
|
+
rules in another `.rules` file, such as `default.rules`. Codex applies the
|
|
239
|
+
strictest rule across all files.
|
|
240
|
+
|
|
241
|
+
## Check the deployed rules
|
|
242
|
+
|
|
243
|
+
Each check below is safe, because the shell never reaches `rm`.
|
|
244
|
+
|
|
245
|
+
- **Codex**: run
|
|
246
|
+
`codex execpolicy check --pretty --rules ~/.codex/rules/docks.rules -- rm -rf /`.
|
|
247
|
+
It prints the decision and runs nothing.
|
|
248
|
+
- **omp and Claude Code**: ask the agent to run `true || rm -rf "$HOME"`. The
|
|
249
|
+
`true ||` short-circuit stops the shell before `rm`. omp reports
|
|
250
|
+
`Blocked by bash pattern: rm * $HOME`, and Claude Code reports a denial.
|
|
251
|
+
|
|
252
|
+
## Sources
|
|
253
|
+
|
|
254
|
+
- omp: `omp://tools/bash.md`, `omp://approval-mode.md`, and the
|
|
255
|
+
`@oh-my-pi/pi-coding-agent` 18.11.1 source `src/tools/bash.ts`
|
|
256
|
+
(`CRITICAL_BASH_PATTERNS`, `BashTool.approval`).
|
|
257
|
+
- Claude Code: https://code.claude.com/docs/en/permissions and
|
|
258
|
+
https://code.claude.com/docs/en/permission-modes
|
|
259
|
+
- Codex: https://learn.chatgpt.com/docs/agent-configuration/rules,
|
|
260
|
+
https://github.com/openai/codex/blob/main/codex-rs/execpolicy/README.md, and
|
|
261
|
+
https://learn.chatgpt.com/docs/config-file/config-reference (`approvals_reviewer`).
|
|
262
|
+
- GLib `gio trash`: https://github.com/GNOME/glib/blob/main/gio/gio-tool-trash.c
|
|
263
|
+
- Git: https://git-scm.com/docs/gitcli (abbreviated and clustered options),
|
|
264
|
+
https://git-scm.com/docs/git-rm, https://git-scm.com/docs/git-worktree,
|
|
265
|
+
https://github.com/git/git/blob/master/builtin/push.c (short options), and
|
|
266
|
+
https://github.com/git/git/blob/master/builtin/worktree.c
|
|
267
|
+
(`remove_worktree` parses `--`; `check_clean_worktree` runs
|
|
268
|
+
`git status --porcelain` without `--ignored`).
|
|
269
|
+
- macOS `trash`: https://keith.github.io/xcode-man-pages/trash.8.html
|
package/cli/docs/install.md
CHANGED
|
@@ -50,8 +50,7 @@ package with the pinned `effect`.
|
|
|
50
50
|
## 3. curl installer (POSIX)
|
|
51
51
|
|
|
52
52
|
```
|
|
53
|
-
curl -fsSL https://raw.githubusercontent.com/DocksDocks/public/main/install.sh -o
|
|
54
|
-
bash /tmp/docks-kit-install.sh && rm /tmp/docks-kit-install.sh
|
|
53
|
+
tmp="$(mktemp)" && curl -fsSL https://raw.githubusercontent.com/DocksDocks/public/main/install.sh -o "$tmp" && bash "$tmp"; rc=$?; rm -f "$tmp"; (exit "$rc")
|
|
55
54
|
```
|
|
56
55
|
|
|
57
56
|
Download-then-run, never `curl | bash` — stream truncation has bitten this
|
|
@@ -70,8 +69,15 @@ Remove-Item "$env:TEMP\docks-kit-install.ps1"
|
|
|
70
69
|
|
|
71
70
|
Download the script before execution. Never stream it into PowerShell. The
|
|
72
71
|
installer bootstraps Bun when absent. It then runs
|
|
73
|
-
`bun add -g docks-kit@latest` and
|
|
74
|
-
`%USERPROFILE%\.
|
|
72
|
+
`bun add -g docks-kit@latest` and runs `docks-kit --version` from Bun's global
|
|
73
|
+
bin directory (`%USERPROFILE%\.bun\bin` by default). The CLI stays in that
|
|
74
|
+
directory, because Bun's `docks-kit.exe` shim cannot run without the
|
|
75
|
+
`docks-kit.bunx` file next to it. When the directory is not on `PATH`, the
|
|
76
|
+
installer prints the command that adds it. The installer removes a copy of
|
|
77
|
+
`docks-kit.exe` that an earlier installer left in `%USERPROFILE%\.local\bin`
|
|
78
|
+
when the copy is identical to the current shim. When another `docks-kit`
|
|
79
|
+
executable earlier on `PATH` shadows Bun's shim, the installer names that file.
|
|
80
|
+
It also prints the command that removes the file and exits with status 1.
|
|
75
81
|
|
|
76
82
|
## Keeping the kit up to date
|
|
77
83
|
|
|
@@ -99,11 +105,13 @@ docks-kit checkout. `DOCKS_KIT_HOME` lets you target a specific checkout on
|
|
|
99
105
|
purpose.
|
|
100
106
|
|
|
101
107
|
Autodetection: a kit home with `.git` is a checkout (requires a clean
|
|
102
|
-
worktree and an upstream; `git pull --ff-only`, re-runs
|
|
103
|
-
`bun install --frozen-lockfile` when the
|
|
104
|
-
under `node_modules` is a global package
|
|
105
|
-
`npm install -g docks-kit@latest`). The chained sync
|
|
106
|
-
|
|
108
|
+
worktree and an upstream; `git pull --ff-only`, then re-runs
|
|
109
|
+
`bun install --frozen-lockfile` when the pull changed `bun.lock` or
|
|
110
|
+
`package.json`); a kit home under `node_modules` is a global package
|
|
111
|
+
(`bun add -g` / `npm install -g docks-kit@latest`). The chained sync is a
|
|
112
|
+
plain flag-less `docks-kit sync`. It refreshes plugin marketplaces and
|
|
113
|
+
installed plugins even when the kit is already current. Pass `--no-sync`, then
|
|
114
|
+
run `docks-kit sync --skip-plugin-refresh` to skip the refresh-only work.
|
|
107
115
|
A compiled binary inside a checkout updates the checkout; on the next invocation
|
|
108
116
|
the launcher bypasses that now-stale binary and uses updated source until rebuilt.
|
|
109
117
|
Every `docks-kit sync` also does a best-effort behind-upstream check and
|
|
@@ -155,7 +163,10 @@ sync/config reads.
|
|
|
155
163
|
- Node/npm for npm-global LSP servers
|
|
156
164
|
- rustup is optional and needed only for the Rust language server
|
|
157
165
|
- jq is optional doctor/test tooling; sync has no jq runtime dependency
|
|
158
|
-
- curl is
|
|
159
|
-
|
|
160
|
-
|
|
166
|
+
- curl is needed only to download Bun's upstream installer when Bun is absent:
|
|
167
|
+
by the POSIX `docks-kit` launcher, by the POSIX `install.sh`, and by sync's
|
|
168
|
+
Bun bootstrap on every host, release binaries included
|
|
169
|
+
(`docks-kit toolchain ensure bun` too). `docks-kit.ps1` and `install.ps1`
|
|
170
|
+
download Bun's installer with Invoke-WebRequest. An already-present Bun needs
|
|
171
|
+
no curl.
|
|
161
172
|
- See `docks-kit toolchain check` for the full picture on this machine
|
package/cli/docs/models.md
CHANGED
|
@@ -4,8 +4,9 @@
|
|
|
4
4
|
enabled harness, `docks-kit models` and the picker list live model IDs when
|
|
5
5
|
available. Claude uses the Claude Code login to query the Anthropic models API
|
|
6
6
|
and caches the result for six hours in `~/.docks-kit/kit.db`; Codex reads
|
|
7
|
-
|
|
8
|
-
omp to list
|
|
7
|
+
`models_cache.json` in the Codex home (`$CODEX_HOME` when set, else
|
|
8
|
+
`~/.codex`); omp runs `omp models --json`. The kit never needs omp to list
|
|
9
|
+
Claude models.
|
|
9
10
|
|
|
10
11
|
Live lists retain the curated aliases and notes, but omit curated IDs absent
|
|
11
12
|
from the live source. A disabled harness, missing login/cache, or unavailable
|
package/cli/docs/modifiers.md
CHANGED
|
@@ -5,9 +5,10 @@ the SoT is never touched. They all share one contract:
|
|
|
5
5
|
|
|
6
6
|
> A later flag-less sync reverts the modifier: the settings merge (Claude)
|
|
7
7
|
> and the config.toml merge (Codex) re-assert SoT values for kit-owned keys.
|
|
8
|
-
> Re-pass the flag on machines that should keep the override. Claude
|
|
9
|
-
>
|
|
10
|
-
>
|
|
8
|
+
> Re-pass the flag on machines that should keep the override. Claude Code has
|
|
9
|
+
> no user-scope local settings file; a per-project Claude override belongs in
|
|
10
|
+
> that checkout's `.claude/settings.local.json`, which sync never reads or
|
|
11
|
+
> writes.
|
|
11
12
|
|
|
12
13
|
Claude's embedded SoT is `model: opus`, `effortLevel: high`, with advisor
|
|
13
14
|
off (`advisorModel` unset). Codex's embedded normal and plan reasoning effort is
|
package/cli/docs/omp-context.md
CHANGED
|
@@ -90,12 +90,19 @@ No current role model has a window below 272,000, so the hazard was latent
|
|
|
90
90
|
rather than live. Not pinning removes it entirely, including for any future
|
|
91
91
|
model with a smaller window.
|
|
92
92
|
|
|
93
|
+
## Idle compaction is a kit-set value
|
|
94
|
+
|
|
95
|
+
`SoT/.omp/config.yml` sets `compaction.idleEnabled: true`. The omp default is
|
|
96
|
+
`false` in omp 18.4.4 through 18.10.4. Idle compaction fires below the
|
|
97
|
+
reserve-based trigger on both lanes. A compaction whose reason is `idle`
|
|
98
|
+
skips the progress guard. Keep the key. The merge is additive, so a dropped
|
|
99
|
+
key turns idle compaction off on fresh installs.
|
|
100
|
+
|
|
93
101
|
## Related settings left at their omp defaults
|
|
94
102
|
|
|
95
103
|
| Setting | Value | Reason |
|
|
96
104
|
|---|---|---|
|
|
97
|
-
| `extendedContext` | `false` | omp 17.4.0 added the setting
|
|
98
|
-
| `compaction.idleEnabled` | `true` | omp default. Idle compaction fires below the reserve-based trigger on both lanes. A compaction whose reason is `idle` skips the progress guard. |
|
|
105
|
+
| `extendedContext` | `false` | omp 17.4.0 added the setting with the default on, and a later release changed the default to off. The setting controls the window omp reports for a model with a long-context limit. On omp 18.10.4, `omp models openai-codex` lists `openai-codex/gpt-6.1-sol` at 500,000 tokens with the setting off and 922,000 tokens with it on. The Codex lane uses this model for the `default`, `task`, and `vision` fallbacks and for the `reviewer` role. OpenAI bills the whole request at the long-prompt rate when the input is above 272,000 tokens. That rate is input $4.00 instead of $2.00, cache read $0.20 instead of $0.10, cache write $5.00 instead of $2.50, and output $15.00 instead of $10.00. The cost is a cliff rather than a marginal rate. Keeping the setting off keeps the smaller window, so fewer requests cross that cliff. |
|
|
99
106
|
| `compaction.idleThresholdTokens` | `200000` | omp default. |
|
|
100
107
|
| `compaction.idleTimeoutSeconds` | `300` | omp default. |
|
|
101
108
|
| `compaction.methodOrder` | `["remote","snapcompact","handoff","shake","soft"]` | omp default, remote first. Since omp 18.1.18 the Anthropic server-side lane handles Anthropic models, so snapcompact is a fallback rather than the primary path. |
|