Updates SECURITY-MODEL.md §3 to reflect #8248, which
registers Ollama, llama.cpp, LocalAI, vLLM, LM Studio, RHAIIS, Groq,
OpenRouter and Open WebUI as chat-only agent providers in
InitializeProviders.

Changes:

- Provider table flips the Registered column from "no" to "yes (chat
  only)" for the nine HTTP providers that are now wired into the agent
  dropdown, and adds rows for the six new local LLM runners with their
  env vars and default URLs.
- Explains the chat-only capability flag and why missions still route
  through the tool-capable CLI agents (registry.go:303 rationale).
- Adds a "Local LLM strategy" subsection that cross-links the
  docs.kubestellar.io local-llm-strategy page and the eight install
  missions on kubestellar/console-kb.
- Replaces the "Planned follow-up" subsection with active recipes for
  each runner — Ollama loopback default, in-cluster Service URLs for
  llama.cpp/LocalAI/vLLM/RHAIIS, LM Studio workstation default, and
  Groq/OpenRouter/Open WebUI gateway overrides. The "# PLANNED —
  not yet wired at runtime" bash comments are removed.

The threat model claims about kubeconfig and credentials staying out
of the request body are unchanged and still authoritative.

Signed-off-by: Andrew Anderson <[email protected]>

…utes (#8252)

* 📝 docs(security): mark local LLM providers registered + active

Updates SECURITY-MODEL.md §3 to reflect #8248, which
registers Ollama, llama.cpp, LocalAI, vLLM, LM Studio, RHAIIS, Groq,
OpenRouter and Open WebUI as chat-only agent providers in
InitializeProviders.

Changes:

- Provider table flips the Registered column from "no" to "yes (chat
  only)" for the nine HTTP providers that are now wired into the agent
  dropdown, and adds rows for the six new local LLM runners with their
  env vars and default URLs.
- Explains the chat-only capability flag and why missions still route
  through the tool-capable CLI agents (registry.go:303 rationale).
- Adds a "Local LLM strategy" subsection that cross-links the
  docs.kubestellar.io local-llm-strategy page and the eight install
  missions on kubestellar/console-kb.
- Replaces the "Planned follow-up" subsection with active recipes for
  each runner — Ollama loopback default, in-cluster Service URLs for
  llama.cpp/LocalAI/vLLM/RHAIIS, LM Studio workstation default, and
  Groq/OpenRouter/Open WebUI gateway overrides. The "# PLANNED —
  not yet wired at runtime" bash comments are removed.

The threat model claims about kubeconfig and credentials staying out
of the request body are unchanged and still authoritative.

Signed-off-by: Andrew Anderson <[email protected]>

* 🔥 chore(ci): delete 10 dead Copilot workflows + reclaim wasted CI minutes

Third of four PRs from the fullsend-ai/fullsend automation
evaluation. The plan was to replace copilot-pr-monitor.yml's minutely
cron poll with webhook triggers — but closer inspection found 10
workflows in the same fully-disabled state, and two of them were
burning CI minutes on unconditional cron schedules:

- copilot-pr-monitor.yml: cron * * * * * → 1440 wasted runs/day
- copilot-retry.yml:       cron */5 * * * * → 288 wasted runs/day

Every job in every deleted workflow is guarded by:
  if: false # Copilot disabled — issues handled by Claude Code scanner

Deleting them rather than refactoring is the right call per fullsend's
"repo-as-coordinator via native primitives" principle — dead code
pretending to orchestrate nothing is worse than no code at all.

Deleted:
- copilot-pr-monitor.yml (minutely cron poll, 100% no-op)
- copilot-retry.yml       (5-min cron, 100% no-op)
- copilot-assigned.yml
- copilot-automation.yml
- copilot-build-check.yml
- copilot-build-monitor.yml
- copilot-dco.yml
- copilot-recovery.yml
- copilot-review-apply.yml
- ai-fix.yml

Kept:
- copilot-comment-followup.yml — active, reads Copilot review comments
  on merged PRs. Unrelated to the deleted pipeline.
- copilot-setup-steps.yml — harmless workflow_dispatch-only helper.

Verification:
- No `workflow_run:` trigger in workflow-failure-issue.yml references
  any of the deleted workflow names.
- No `uses: ./.github/workflows/...` references to any of the deleted
  files.
- The `ai-fix-requested` label name is unrelated to `ai-fix.yml` —
  label stays, workflow file goes.

Not a replacement for the Claude Code scanner approach — that's
already running and owns the issue-fix loop now. This PR is pure
cleanup.

Savings: ~1728 dead workflow runs per day, plus a lot less noise in
the Actions tab.

Signed-off-by: Andrew Anderson <[email protected]>

---------

Signed-off-by: Andrew Anderson <[email protected]>

…red output (#8253)

* 📝 docs(security): mark local LLM providers registered + active

Updates SECURITY-MODEL.md §3 to reflect #8248, which
registers Ollama, llama.cpp, LocalAI, vLLM, LM Studio, RHAIIS, Groq,
OpenRouter and Open WebUI as chat-only agent providers in
InitializeProviders.

Changes:

- Provider table flips the Registered column from "no" to "yes (chat
  only)" for the nine HTTP providers that are now wired into the agent
  dropdown, and adds rows for the six new local LLM runners with their
  env vars and default URLs.
- Explains the chat-only capability flag and why missions still route
  through the tool-capable CLI agents (registry.go:303 rationale).
- Adds a "Local LLM strategy" subsection that cross-links the
  docs.kubestellar.io local-llm-strategy page and the eight install
  missions on kubestellar/console-kb.
- Replaces the "Planned follow-up" subsection with active recipes for
  each runner — Ollama loopback default, in-cluster Service URLs for
  llama.cpp/LocalAI/vLLM/RHAIIS, LM Studio workstation default, and
  Groq/OpenRouter/Open WebUI gateway overrides. The "# PLANNED —
  not yet wired at runtime" bash comments are removed.

The threat model claims about kubeconfig and credentials staying out
of the request body are unchanged and still authoritative.

Signed-off-by: Andrew Anderson <[email protected]>

* ✨ feat(ci): decomposed-review prompt for Claude Code review — structured output

Fourth of four PRs from the fullsend-ai/fullsend automation
evaluation. Adopts fullsend's "decomposed code review" pattern
(docs/problems/code-review.md) — split review into specialized
concerns (correctness, security, style) rather than one monolithic
pass.

Done as a single LLM call with structured output, not 3 parallel
jobs, to keep token cost neutral. The prompt asks the existing
/code-review:code-review plugin to organize its findings into three
explicit sections with P0/P1/P2 priority tags, and to write "None."
under any section that has nothing to report so it doesn't
fabricate issues.

The SECURITY section references docs/security/SECURITY-AI.md (added
in PR #8249) so the reviewer explicitly watches for the six threat
categories — external prompt injection, insider credentials, DoS,
agent drift, supply chain, agent-to-agent injection — on any PR
that touches LLM-calling code.

Only change is the `prompt:` field in the existing
`.github/workflows/claude-code-review.yml`. No new actions, no new
secrets, no cost increase.

Expected behavior after merge:
- Every PR's Claude Code review comment now has a CORRECTNESS /
  SECURITY / STYLE structure instead of prose.
- Every issue is tagged P0/P1/P2 so reviewers can triage quickly.
- A pure doc PR should show "None." in all three sections, not
  fabricated nits.
- A PR touching LLM-calling code should produce at least one item
  in SECURITY referencing prompt-injection risk.

Signed-off-by: Andrew Anderson <[email protected]>

---------

Signed-off-by: Andrew Anderson <[email protected]>

Updates SECURITY-MODEL.md §3 to reflect #8248, which
registers Ollama, llama.cpp, LocalAI, vLLM, LM Studio, RHAIIS, Groq,
OpenRouter and Open WebUI as chat-only agent providers in
InitializeProviders.

Changes:

- Provider table flips the Registered column from "no" to "yes (chat
  only)" for the nine HTTP providers that are now wired into the agent
  dropdown, and adds rows for the six new local LLM runners with their
  env vars and default URLs.
- Explains the chat-only capability flag and why missions still route
  through the tool-capable CLI agents (registry.go:303 rationale).
- Adds a "Local LLM strategy" subsection that cross-links the
  docs.kubestellar.io local-llm-strategy page and the eight install
  missions on kubestellar/console-kb.
- Replaces the "Planned follow-up" subsection with active recipes for
  each runner — Ollama loopback default, in-cluster Service URLs for
  llama.cpp/LocalAI/vLLM/RHAIIS, LM Studio workstation default, and
  Groq/OpenRouter/Open WebUI gateway overrides. The "# PLANNED —
  not yet wired at runtime" bash comments are removed.

The threat model claims about kubeconfig and credentials staying out
of the request body are unchanged and still authoritative.

Signed-off-by: Andrew Anderson <[email protected]>

…RL Advanced section

Part 2 of #8254. The Settings → API Keys modal now renders the nine
chat-only providers (Groq, OpenRouter, Open WebUI + Ollama, llama.cpp,
LocalAI, vLLM, LM Studio, RHAIIS) with an Advanced expandable section
that accepts a per-provider Base URL override.

Backend changes
---------------

- KeyStatus gains BaseURL, BaseURLEnvVar, BaseURLSource fields so the
  frontend can render the current resolved value and indicate whether
  it came from env or config file.
- handleGetKeysStatus now returns the nine chat-only providers (was
  previously an empty list per the "API-key-driven agents are hidden"
  comment, which predated PR #8248 flipping them on). CLI-based
  tool-capable agents (claude-code, bob, codex, ...) remain omitted —
  they manage their own credentials.
- SetKeyRequest.APIKey is now optional: setting just BaseURL is the
  common path for unauthenticated local LLM runners. handleSetKey
  accepts any combination of apiKey, baseURL, model — at least one
  is required. Base URL is syntactically validated (http(s)://, no
  whitespace) via validateBaseURL before being saved. On successful
  save, cached key validity is invalidated so the next validation
  hits the new endpoint.
- Key validation is skipped for local LLM runners (validationRequired
  is false on their providerDef entries). There is no meaningful way
  to validate the sentinel "local-llm-no-auth" placeholder against a
  real endpoint, so we report Configured=true based on URL presence.

Frontend changes
----------------

- KeyStatus TS interface gains baseURL, baseURLEnvVar, baseURLSource.
- APIKeySettings.tsx adds an Advanced expandable to each row. It
  only renders when the backend reports baseURLEnvVar (meaning the
  provider actually supports an override). The form includes a
  text input, an inline "env var wins" hint when baseURLSource is
  "env", syntactic validation feedback, and a post-save "restart
  kc-agent to apply" message. The summary row shows the current
  resolved value next to the Advanced toggle so operators can see
  it at a glance.

Closes #8254.

- go build ./... passes.
- go test ./pkg/agent/... passes (flaky auto-update test unrelated).
- npm run build passes all post-build safety checks.

Signed-off-by: Andrew Anderson <[email protected]>

…closes #8254) (#8256)

* ✨ feat(agent): ConfigManager.GetBaseURL/SetBaseURL + dynamic provider base URL resolution

Part 1 of #8254. Adds per-provider base URL storage in ~/.kc/config.yaml
and updates all nine chat-only HTTP providers to resolve their base URL
dynamically at request time, so changes take effect without restarting
kc-agent.

Backend changes
---------------

- AgentKeyConfig gains an optional `base_url` YAML field alongside
  api_key and model. Empty string means "use the compiled-in default".
- ConfigManager.GetBaseURL(provider) / SetBaseURL / RemoveBaseURL
  mirror the existing GetAPIKey / SetAPIKey / RemoveAPIKey shape. The
  precedence chain is env var → config file → compiled default, same
  as GetAPIKey and GetModel.
- getBaseURLEnvKeyForProvider maps each registered provider key to
  its URL env var (OLLAMA_URL, LLAMACPP_URL, LOCALAI_URL, VLLM_URL,
  LM_STUDIO_URL, RHAIIS_URL, GROQ_BASE_URL, OPENROUTER_BASE_URL,
  OPEN_WEBUI_URL).

Provider refactor
-----------------

- LocalOpenAICompatProvider.localOpenAICompatBaseURL() now walks
  env var → ConfigManager.GetBaseURL → struct default. All six local
  LLM runners use this path.
- GroqProvider drops its cached baseURL field and resolves via a new
  groqResolveBaseURL() helper (env → config → default). The existing
  groqValidationURL() already consults the same helper, so key
  validation continues to honor the override.
- OpenRouterProvider and OpenWebUIProvider get the same treatment —
  dynamic resolution via openRouterResolveBaseURL() and
  openWebUIResolveBaseURL() respectively.

The frontend UI (APIKeySettings Advanced section with a Base URL
input field per provider) is Part 2 of this PR, coming next.

- go build ./... passes.
- go test ./pkg/agent/... passes.

Signed-off-by: Andrew Anderson <[email protected]>

* ✨ feat(ui): Settings → API Keys lists local LLM providers with Base URL Advanced section

Part 2 of #8254. The Settings → API Keys modal now renders the nine
chat-only providers (Groq, OpenRouter, Open WebUI + Ollama, llama.cpp,
LocalAI, vLLM, LM Studio, RHAIIS) with an Advanced expandable section
that accepts a per-provider Base URL override.

Backend changes
---------------

- KeyStatus gains BaseURL, BaseURLEnvVar, BaseURLSource fields so the
  frontend can render the current resolved value and indicate whether
  it came from env or config file.
- handleGetKeysStatus now returns the nine chat-only providers (was
  previously an empty list per the "API-key-driven agents are hidden"
  comment, which predated PR #8248 flipping them on). CLI-based
  tool-capable agents (claude-code, bob, codex, ...) remain omitted —
  they manage their own credentials.
- SetKeyRequest.APIKey is now optional: setting just BaseURL is the
  common path for unauthenticated local LLM runners. handleSetKey
  accepts any combination of apiKey, baseURL, model — at least one
  is required. Base URL is syntactically validated (http(s)://, no
  whitespace) via validateBaseURL before being saved. On successful
  save, cached key validity is invalidated so the next validation
  hits the new endpoint.
- Key validation is skipped for local LLM runners (validationRequired
  is false on their providerDef entries). There is no meaningful way
  to validate the sentinel "local-llm-no-auth" placeholder against a
  real endpoint, so we report Configured=true based on URL presence.

Frontend changes
----------------

- KeyStatus TS interface gains baseURL, baseURLEnvVar, baseURLSource.
- APIKeySettings.tsx adds an Advanced expandable to each row. It
  only renders when the backend reports baseURLEnvVar (meaning the
  provider actually supports an override). The form includes a
  text input, an inline "env var wins" hint when baseURLSource is
  "env", syntactic validation feedback, and a post-save "restart
  kc-agent to apply" message. The summary row shows the current
  resolved value next to the Advanced toggle so operators can see
  it at a glance.

Closes #8254.

- go build ./... passes.
- go test ./pkg/agent/... passes (flaky auto-update test unrelated).
- npm run build passes all post-build safety checks.

Signed-off-by: Andrew Anderson <[email protected]>

---------

Signed-off-by: Andrew Anderson <[email protected]>

…8314)

The nightly Release workflow has been red since 2026-04-13 because four
Go tests fell behind handler changes shipped over the past two weeks.
Each test now aligns with the current handler contract:

- `pkg/agent/server_test.go:TestServer_SettingsHandlers` — expected the
  keys endpoint to return 0 providers, but #8248/#8254/#8256 now
  register nine chat-only HTTP providers (3 OpenAI-compatible gateways
  + 6 local LLM runners) so the Settings modal can show per-provider
  base URL overrides. Assert non-empty list and Provider presence.

- `pkg/api/handlers/cards_test.go:TestRecordFocus_BadBody_Returns400` —
  `RecordFocus` now runs `requireEditorOrAdmin` before BodyParser (#7011),
  which calls `store.GetUser`. The test's custom `recordFocusStore`
  never registered `GetUser`, so the mock panicked. Add the admin-user
  expectation.

- `pkg/api/handlers/dashboard_test.go:setupDashboardTest` —
  `CreateDashboard` now calls `store.CountUserDashboards` to enforce
  the per-user dashboard limit (#7010). Register a default
  `Return(0, nil).Maybe()` so the three CreateDashboard tests stay
  under the limit. Tests exercising the limit can override.

- `pkg/api/handlers/setup_test.go` — `TestClusterGroupsCRUD` exercises
  handlers that persist definitions via `SaveClusterGroup` /
  `DeleteClusterGroup` / `ListClusterGroups` (#7013). Register
  permissive `.Maybe()` mocks on the shared `setupTestEnv` MockStore.

Verification: `go test ./...` — all packages pass locally.

Fixes #8310, Fixes #8313, Fixes #8314

Signed-off-by: Andy Anderson <[email protected]>

…8314) (#8317)

The nightly Release workflow has been red since 2026-04-13 because four
Go tests fell behind handler changes shipped over the past two weeks.
Each test now aligns with the current handler contract:

- `pkg/agent/server_test.go:TestServer_SettingsHandlers` — expected the
  keys endpoint to return 0 providers, but #8248/#8254/#8256 now
  register nine chat-only HTTP providers (3 OpenAI-compatible gateways
  + 6 local LLM runners) so the Settings modal can show per-provider
  base URL overrides. Assert non-empty list and Provider presence.

- `pkg/api/handlers/cards_test.go:TestRecordFocus_BadBody_Returns400` —
  `RecordFocus` now runs `requireEditorOrAdmin` before BodyParser (#7011),
  which calls `store.GetUser`. The test's custom `recordFocusStore`
  never registered `GetUser`, so the mock panicked. Add the admin-user
  expectation.

- `pkg/api/handlers/dashboard_test.go:setupDashboardTest` —
  `CreateDashboard` now calls `store.CountUserDashboards` to enforce
  the per-user dashboard limit (#7010). Register a default
  `Return(0, nil).Maybe()` so the three CreateDashboard tests stay
  under the limit. Tests exercising the limit can override.

- `pkg/api/handlers/setup_test.go` — `TestClusterGroupsCRUD` exercises
  handlers that persist definitions via `SaveClusterGroup` /
  `DeleteClusterGroup` / `ListClusterGroups` (#7013). Register
  permissive `.Maybe()` mocks on the shared `setupTestEnv` MockStore.

Verification: `go test ./...` — all packages pass locally.

Fixes #8310, Fixes #8313, Fixes #8314

Signed-off-by: Andy Anderson <[email protected]>