Two enhancements requested after live-validating PR #430 against a real
llama.cpp server:
1. Model picker dialog. Picking a Run-menu Custom Endpoints entry used to
apply the endpoint's defaultModelId (or the first discovered model)
silently. Now, via the new selectCustomModelEntry() (session-ui.js):
- exactly one discovered model launches straight away, same as before
- two or more open a new #customModelPickModal listing every discovered
model; defaultModelId (if set) is marked but never auto-chosen, since
the point of asking is letting ONE launch deliberately differ from
the saved default, not just confirming it
The endpoint is re-fetched at click time rather than trusting anything
cached from the dropdown's own render, since the model list can have
changed (the sweep below, or a settings-panel edit) since it opened.
runCustomModelEntry() itself — the actual launch, routed through run()
for the in-flight lock, snapshot-guarded against applying to the wrong
session — is unchanged; it now just always receives an explicit model
id from one of these two paths instead of computing one itself.
2. Periodic re-discovery. Every saved endpoint's models now refresh
automatically every 5 minutes in the background
(CUSTOM_MODEL_REDISCOVER_INTERVAL_MS, server.ts, registered the same way
as the Codex plan-usage poll it sits beside — this.cleanup.setInterval,
off under testMode), so a model the server starts or stops serving shows
up without another manual "Discover" click. The manual POST
.../discover-models route and the new refreshAllCustomModelHosts()
sweep (custom-model-routes.ts) now share one pure merge step
(applyDiscoveredModels: stamps lastDiscoveredAt, drops a defaultModelId
that no longer appears) rather than two copies that could drift. The
sweep is best-effort per host — one endpoint being unreachable on a
cycle never blocks the others — and re-reads the store before each
host's write, keyed by id, so a concurrent edit or delete from the
settings panel always wins over a sweep that started before it.
Tests: test/custom-model-endpoint-rediscovery.test.ts is a new, dedicated
file for the sweep (kept separate from custom-model-routes.test.ts because
that file's data dir is shared across every test in it — one temp HOME per
FILE, not per test — which would make a sweep-touches-every-host assertion
meaningless there). test/custom-model-run-menu-ui.test.ts gained a new
describe block driving the real picker modal through JSDOM: single-model
bypass, multi-model dialog with the default marked-not-chosen, picking a
row closes the modal and launches with that exact model, the endpoint
re-fetch, and the two "vanished by click time" toast paths.
Docs: docs/custom-model-endpoints.md, docs/wiki/Custom-Model-Endpoints.md,
docs/api-reference.md and CLAUDE.md's dense feature paragraph all updated
— the last of these also caught up two sentences that had gone stale after
the draft-review fixes landed (the picker routes through run() now, not a
raw run*() call).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RqZeHrRS6DYcGcGX2p9EwG
Point any Codeman-supported harness (Claude, opencode, Codex, Gemini, Pi,
Grok, DeepSeek, OMP) at a custom OpenAI-compatible endpoint instead of its
native cloud backend, for a given session. Covers local hardware (llama.cpp,
Ollama, vLLM, DGX Spark, Strix Halo) and cloud (Azure AI Foundry, OpenRouter).
Off by default (customModelEndpointsEnabled, synced, default OFF).
- Registry: capabilities.customModelInjection per CLI entry (env /
configContentEnv / configDir / unsupported kinds)
- Pure injection builder (custom-model-injection.ts) turning an endpoint +
model id into the real env vars / config content per CLI
- Endpoint store + CRUD routes (custom-model-hosts.ts,
custom-model-routes.ts), discovery via GET /v1/models, SSRF-guarded
- Session integration: Session.setCustomModel()/restartCli()
(POST /api/sessions/:id/custom-model), reusing the existing
respawn-pane -k primitive to restart the CLI process with new env
- Multi-user hardening: every new redirect-capable env var added to its
CLI's privilegedEnvKeys, closing a pre-existing gap where several were
already reachable via the generic envOverrides field's prefix allowlist
- Standalone scripts/test-local-llm-harnesses.mjs: spawns real CLI binaries
against a real endpoint outside the web UI, independent of tmux/sessions
- Mock-server contract tests (test/fixtures/mock-openai-server.ts) replaying
every CLI's injected values through a real HTTP shape
Real end-to-end validation against a live llama-swap server (inside a
codeman/agent:llm-test Docker image with all 9 CLI binaries) found and
fixed three real bugs before they shipped:
- Codex's config.toml schema was wrong ([model].default table instead of
a top-level model string + [model_providers.custom]); fixing it then
surfaced a genuine, documented protocol incompatibility (Codex only
speaks the Responses API since Feb 2026, which llama.cpp/llama-swap
don't implement)
- Claude Code's async session-title-generation call validates
ANTHROPIC_DEFAULT_HAIKU_MODEL against its own internal model list and
hangs the whole -p invocation on an unrecognized name; documented for
chunk 6, worked around in the standalone script only (--bare is NOT
safe for a real interactive session, which needs hooks)
- The discovery route's authStyle: 'both' option (send both Authorization
and api-key headers) reliably hung a real server; removed the option
entirely rather than just changing the default
Status: draft. Chunk 6 (frontend toolbar/settings UI) not yet built — see
PR.md and deployment_plan.md for the full chunk breakdown and confidence
table.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017HqNWfmtBU2KN29SvSVWB3
The mic button previously needed a Deepgram API key, or fell back to the
browser's Web Speech engine. It can now transcribe through the same
speech-to-text service Claude Code's own /voice mode uses, so anyone signed
in to Claude Code on the server gets dictation with no third-party account.
Claude Code's voice mode cannot be driven directly: it opens the HOST's
microphone (sox/arecord), and the CLI runs in a headless tmux pane while the
human is in a browser somewhere else. So capture stays in the browser and only
the transcription backend is borrowed.
Audio goes browser -> Codeman -> Anthropic. The OAuth token never reaches the
page: the browser sends PCM16 (16 kHz mono, produced by an AudioWorklet since
MediaRecorder cannot emit raw PCM) and receives text.
- GET /api/voice/status reports readiness and never the token
- GET /ws/voice/stream relays one dictation, with the same Host/Origin upgrade
guard as the terminal socket, plus caps on concurrency, stream length and
frame size
- credentials are read-only: Codeman never refreshes them, since a refresh
rotates the refresh token and could sign the user out of their own CLI
- claudeVoiceEnabled (synced, default OFF) gates the whole server side
- voiceSettings.provider picks auto/claude/deepgram/webspeech; auto prefers
Claude, then a configured Deepgram key, then the browser
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Per-case profiles of user intent (docs/readmymind-plan.md): user-stated goals
plus the user's recently submitted prompts, captured from the Claude session
transcript behind the new synced readMyMindEnabled setting (default OFF).
- intent-store.ts: keyed by owner + realpath(workingDir), FIFO/size caps,
consecutive-dupe collapse, atomic 0600 writes to ~/.codeman/intents.json
- transcript-watcher.ts: new transcript:user_prompt event for typed user turns
(tool_result-only entries stay silent); capture wiring in server.ts is
claude-only and gated on the setting per event
- readmymind-routes.ts: GET/PUT/DELETE /api/sessions/:id/intent, ownership
via findSessionOrFail, strict Zod schema
- agent skill: SKILL.md recipe + endpoints.md rows so agents can read and
record intent (PUT replaces: read + merge; never delete unprompted)
- groundwork for the phase-2 predictor button; nothing is ever auto-sent
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Permission dialogs, AskUserQuestion questions and idle prompts from every
session now land in a server-side inbox (web/approval-inbox.ts, one item per
session, claude-mode only) and are answerable in place: a header bell + drawer
on desktop, inline answer strips on the phone overview's NEEDS YOU rows, and
working push Approve/Deny buttons (previously dead ends, now answered straight
from sw.js with no tab open). Pending alerts survive reloads because the
frontend seeds from GET /api/approvals on init.
Answering sends the digit / Esc / prompt text through the existing tmux input
path; option digits are accepted only when they match options parsed from the
captured pane frame, and the answer path re-captures the pane first so a
dialog that already left the screen refuses with 409 instead of typing into
the composer. New elicitation_complete / elicitation_response hook matchers
resolve question items the moment they are answered in the terminal;
refreshStaleCodemanHooks heals existing cases.
Verified end-to-end against a live claude session: a real AskUserQuestion
dialog parsed into 5 option buttons and was answered from the drawer.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Adds a "Web / URL" section to the Run dropdown. A saved URL renders as a tab in
the same strip as Claude/Codex/Gemini sessions, with the same Alt+1..9 numbering,
so Codeman is one mission control instead of Codeman plus a pile of browser tabs.
A webview is NOT a sixth SessionMode: no PTY, no tmux, no respawn, no idle
detection. It is a separate resource sharing only the tab strip and the main
content area, the same call that keeps Docker and remote-SSH as case overlays.
Dashboards are proxied through Codeman's own origin, because a direct iframe
fails three ways at once in the shipped deployment: prod serves HTTPS behind
tailscale serve, so http:// targets are hard-blocked as mixed content (with no
override at all on iOS Safari); Grafana/Portainer-class dashboards send
X-Frame-Options: DENY; and our own default-src 'self' CSP blocks cross-origin
frames. Proxying dissolves all three and leaves the production CSP byte-for-byte
unchanged, since /webview/... is already covered by 'self'. A useful side effect:
the fetch happens server-side, so a tailnet-only dashboard is reachable from a
phone that is not on the tailnet.
The proxy is not an API surface. It authenticates on a 192-bit capability in the
path (memory-only, rolling TTL, bound to the minting user, revoked on edit or
delete) and is correspondingly exempt from the cookie and Origin checks, because
a sandboxed iframe is opaque-origin: it sends no SameSite=lax cookie and its
writes arrive with Origin: null. The Host allowlist is never bypassed. A second
Referer-keyed form of the exemption exists for root-absolute assets and is fenced
to safe methods on non-/api, non-/ws, non-/q paths.
Iframes omit allow-same-origin unless a URL is explicitly marked trusted, since a
proxied page is served from Codeman's own origin and could otherwise read this
document and drive the agent-spawning API. Authorization and codeman_session are
stripped upstream in BOTH modes, so CODEMAN_PASSWORD cannot leak into a dashboard.
Two things only a real browser reveals, both presenting as the dashboard's own
"Failed to fetch" while the page itself renders fine:
- Runtime-built root-absolute URLs (fetch('/api/data')) escape <base href> and
land on Codeman's root. Widening the Referer fallback into /api would trade
security for it, so an injected shim patches fetch/XHR/WebSocket/EventSource
inside the frame instead, removing the class rather than the guard.
- An opaque-origin document CORS-checks every request, including to the host it
was served from. Script/css/img loads are not CORS-checked, which is why the
page renders while its API calls die. The proxy now emits CORS headers and
answers preflights itself. registerSecurityHeaders answered every OPTIONS with
a bare 204 before routing, carrying no ACAO for Origin: null, so that
short-circuit now exempts a valid capability.
Neither is reproducible with curl, which does not enforce CORS.
Also fixes a pre-existing bug found on the way: .toolbar has backdrop-filter,
making it a stacking context that trapped .run-mode-menu's z-index:1000, so
.welcome-overlay painted over the whole Run menu. With no session open, every
item in it (Claude Code included) was unclickable.
Verified end to end against a real tailnet dashboard: live data, WebSocket push,
no failed requests, and switching tabs does not reload the frame. 98 new tests
cover the pure rewrite helpers, the CORS helper, the shim's rewrite logic, route
CRUD, and every edge of the auth exemption.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
- routes/admin-routes.ts: GET/POST /api/admin/users, PATCH/DELETE
/api/admin/users/:username, reset-password, logout. Multi-user only (404
otherwise), requireAdmin, last-admin invariants, one-time-password on create /
reset (returned once + mustChangePassword), disable/reset/delete revoke cookie
sessions, delete kills the user's live sessions first (normal teardown) and can
delete their space (guarded). Per-user stats (live/active sessions, case count).
- web/admin-audit.ts: append-only ~/.codeman/admin-audit.jsonl (timestamp, acting
admin, action, target, IP) for every user-management action.
- SSE admin:usersChanged + auth:passwordChangeRequired (sse-events.ts + constants.js).
fix(user-store): serialize users.json read-modify-write
touchLastLogin fires on every Basic auth (fire-and-forget) and was racing route
writes (create/update), clobbering records — a real corruption bug surfaced by
the admin tests. All mutators now run under a single write lock, and
touchLastLogin is throttled to once/minute per user to bound disk churn.
Tests: test/admin-routes.test.ts (8, live server) + user-store lock verified by
the existing user-store suite.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Adds a parallel multi-user auth branch (the single-user Basic-auth path is
left byte-identical). Off unless CODEMAN_MULTIUSER/--multiuser.
- middleware/auth.ts: mode-selecting registerAuthMiddleware. New async
multi-user hook verifies username:password against the user store (scrypt),
mints identity-carrying cookies, decorates req.authUser, enforces a per-IP
AND per-username failure bucket, and the mustChangePassword lockbox. The
hook-secret loopback bypass is now a single shared helper used by both
branches. FastifyRequest.authUser module augmentation.
- ports/auth-port.ts: AuthSessionRecord gains username/role/mustChangePassword.
- user-store.ts: verifyPassword (timing-equalized against user enumeration).
- route-helpers.ts: getAuthUser (synthetic admin fallback), canAccessOwned,
requireAdmin, revokeUserSessions; findSessionOrFail gains an optional req for
a NOT_FOUND owner check (dormant until phase 3 wires callers).
- routes/me-routes.ts: GET /api/me (synthetic admin in single-user) and
POST /api/me/password (verify current, min 8, clear mustChangePassword,
revoke other sessions).
- QR: QrTokenRecord + AuthSessionRecord carry a username; tunnel-manager
mintUserToken / consumeTokenWithIdentity / getQrSvgForCode; /q/:code binds
the cookie to the token's user (rejects identity-less tokens in multi-user);
GET /api/tunnel/qr mints a per-user token. Single-user keeps the rotating token.
- server.ts: bootstrap the initial admin from CODEMAN_USERNAME/PASSWORD on first
boot (refuse to start with no users); multi-user with >= 1 user satisfies the
non-loopback auth requirement and the tunnel-enable guard; userFailures bucket
disposal.
- types/api.ts: FORBIDDEN, PASSWORD_CHANGE_REQUIRED, USER_EXISTS, USER_NOT_FOUND,
LAST_ADMIN error codes (message + status wired).
- Session.owner field + getter/setter, SessionState.owner, MuxSession.owner,
CreateSessionOptions.owner (foundation for phase 3 ownership threading).
Tests: test/multiuser-auth.test.ts (10, live server on 3170/3171). Existing auth
suite (auth-security, qr-auth, cod54-hook-event, network-auth-policy) unchanged
and green; full test:ci sweep passes (3519 tests).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Rename the recurring-jobs feature scheduler->cron to disambiguate from the
legacy ScheduledRun system (/api/scheduled), which is left untouched:
- ScheduledJob->CronJob, SchedulerService->CronService
- /api/scheduler/jobs -> /api/cron/jobs; SSE scheduler:* -> cron:*
- state keys cronJobs/cronJobRuns
- files moved to src/cron/, cron-routes.ts, cron-port.ts, types/cron.ts
- frontend cron-ui.js, #cronModal, menu "Cron"
- docs moved to docs/cron-discovery.md + docs/cron-build-brief.md, README guides
- new tests: cron-service.test.ts, cron-time.test.ts
Green: tsc, lint, frontend-syntax, format, 30 cron + 9 legacy scheduled-runs tests.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PmvZR12aX2v8K7YhqxPUAU
Adds a saved/named scheduling layer on top of Codeman's existing session
primitives. Distinct from the legacy run-now ScheduledRun concept.
- types/scheduler.ts: ScheduledJob + ScheduledJobRun
- state-store: persist scheduledJobs/scheduledJobRuns in ~/.codeman/state.json
- scheduler/scheduler-time.ts: pure once/interval/daily/weekly next-run math
- scheduler/scheduler-service.ts: CRUD, Run Now, due-checker tick, run history;
reuses SessionPort (create -> start -> writeViaMux) for launches
- web/routes/scheduler-routes.ts: /api/scheduler/jobs CRUD + run + history
- web/schemas.ts: zod validation with schedule-type-aware refinements
- web/sse-events.ts: scheduler:* events
- server.ts: wire service into route context + 30s background tick loop
- test/scheduler-time.test.ts: 14 unit tests for next-run calculations
Phase 1 discovery recorded in SCHEDULER_DISCOVERY.md.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Rp7JhmQXcYJhmxFMdZuuah
Surface Claude subscription plan usage limits (5-hour rolling + 7-day
weekly: percent used + reset time) in the header, opt-in via App Settings
→ Display → "Plan Usage Limits" (default OFF, no behavior change when off).
A Codeman-managed Claude statusLine exporter forwards the rate_limits JSON
to a new auth-exempt POST /api/status-telemetry (same loopback + hook-secret
gate as /api/hook-event); parsed telemetry broadcasts over SSE
session:statusTelemetry to a header chip (amber >=80%, red >=95%, reset
times on hover). The exporter prints the same summary back as the
in-terminal footer (print-through).
- src/usage-telemetry.ts: pure parser/formatter (epoch-sec -> ms, clamp,
change signature) + test/usage-telemetry.test.ts
- hooks-config.ts: generateStatusLineCommand + applyStatusLineConfig
(add/remove; never clobbers a user's own statusLine)
- session-routes.ts: inject gate (Claude-only, Codeman-managed cases),
driven by create-payload statusLineTelemetry (session-ui.js)
- schemas.ts: StatusTelemetrySchema + showPlanUsageLimits + payload field
- frontend: header chip, applyHeaderVisibilitySettings toggle,
renderIndexHtml strip, _onSessionStatusTelemetry handler
Schema empirically confirmed against Claude Code 2.1.177 (Claude Max):
only five_hour/seven_day windows exist (no Opus-weekly field); rate_limits
is absent before the first API response and for non-subscriber auth. Design
+ verification method in docs/usage-limits-display-plan.md.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
POST /api/clipboard with { text } broadcasts to all connected browsers
via SSE. Browser attempts navigator.clipboard.writeText, falling back
to a modal with manual copy button if blocked.
Enables remote clipboard workflows: CLI tools can push text to the
user's browser clipboard across the network.
Adds a new autonomous loop that accepts high-level goals, generates
phased execution plans via AI, and executes them step-by-step with
verification gates between phases.
Core components:
- OrchestratorLoop: state machine (idle→planning→approval→executing→verifying→completed)
- OrchestratorPlanner: plan generation via PlanOrchestrator, Kahn's algorithm phase grouping
- OrchestratorVerifier: phase verification (strict/moderate/lenient modes)
- Prompt templates for phase execution, team delegation, verification, replanning
API (10 endpoints):
- POST start/approve/reject/pause/resume/stop
- GET status/plan
- POST phase/:id/skip, phase/:id/retry
Frontend: orchestrator-panel.js with SSE-driven state, phase progress, task tracking
Tests: 22 tests (18 route + 4 unit), all passing. Typecheck/lint/format clean.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace per-keystroke HTTP POST + SSE terminal output with a single
bidirectional WebSocket connection for dramatically lower input latency.
The existing SSE+POST paths remain fully functional as fallback.
Server-side: ws-routes.ts provides /ws/sessions/:id/terminal with 8ms
micro-batching and 16KB flush threshold. Each batch is wrapped in
DEC 2026 synchronized update markers so xterm.js renders atomically —
Ink's DA capability negotiation fails through the PTY→server→WS proxy
chain, so without server-injected markers, cursor-up redraws flicker.
Frontend: _connectWs/_disconnectWs manage per-session WS lifecycle.
Input and resize use WS fast path with HTTP POST fallback. SSE terminal
events are suppressed when WS is active to prevent double rendering.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Split 6,710-line server.ts into focused route modules using port
interfaces for dependency injection. 107/109 routes extracted into
12 domain files with auth middleware, 5 port interfaces, and shared
helpers. Server.ts retains orchestration (SSE, lifecycle, state).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
