Tool-calling rounds emit reasoning into the OpenAI-protocol
reasoning_content field rather than visible content, so AgentStep.Thought
is often empty in DB. The history reconstructor only read step.thought,
which made the historical step card silent for those rounds even though
the user saw the reasoning live.
Fall back to step.reasoning_content when step.thought is empty so
reload mirrors the live experience.
MiMo and DeepSeek V3.2/V4 reject multi-turn requests in thinking mode
when the prior assistant message lacks reasoning_content with HTTP 400:
"The reasoning_content in the thinking mode must be passed back to the API."
Agent's ReAct loop is the worst-case scenario — every round produces tool
calls, exactly the case DeepSeek's docs specify reasoning_content MUST
participate in subsequent context.
Plumb reasoning_content through the full assistant-message round-trip:
- chat.Message / types.ChatResponse / types.AgentStep gain a
reasoning_content field (AgentStep persists via the existing
Message.AgentSteps jsonb column, no migration needed).
- streamLLMToEventBus accumulates reasoning chunks into
result.ReasoningContent and surfaces it on the round's ChatResponse.
- engine.runReActIteration writes it onto AgentStep so cross-turn replay
preserves it.
- observe.appendToolResults attaches it to the same-turn assistant
message; agent_history.buildAssistantHistoryMessages does the same on
cross-turn replay.
- RemoteAPIChat.ConvertMessages forwards it on assistant turns to
openai.ChatCompletionMessage.ReasoningContent (already supported by
go-openai); providers that don't recognize the field ignore it.
Tests cover the three boundaries: ConvertMessages serializes it for
upstream, appendToolResults preserves it within the same turn, and
buildAssistantHistoryMessages replays it across turns.
Scope is intentionally limited to Agent mode — KnowledgeQA's chat
pipeline and Anthropic's signed thinking_blocks are separate fixes that
require schema changes (rendered_content / thinking_blocks columns).
Fixes#1302
The Wiki browser exposes a page-link graph (references between Wiki pages)
while KB Settings → Knowledge Graph configures an LLM-extracted
entity-relationship graph. Both are labeled "图谱"/"Graph", which is easy
to confuse.
- Expand graphSettings.description to call out the distinction
- Add tabGraphTip tooltip on the Wiki "图谱" breadcrumb tab
- Extend wikiBrowser.graphNoData empty state with the same clarification
- Update zh-CN, en-US, ko-KR, ru-RU consistently
The "启用实体关系提取" switch in GraphSettings was silently dropped on
save: buildSubmitData only attached extract_config to the request when
both indexingStrategy.graphEnabled AND nodeExtractConfig.enabled were
true. When the graph indexing strategy was off (the default), the
toggle change never reached the backend and reloading the KB always
showed the switch as off.
Always emit extract_config based on the user's actual toggle state so
the value round-trips correctly through updateKBConfig regardless of
the indexing strategy selection.
OpenAI's GPT-5 series and o-series reasoning models (o1/o3/o4-mini) no
longer accept `max_tokens` and reject non-default sampling params
(`temperature`, `top_p`, `frequency_penalty`, `presence_penalty`).
Azure OpenAI propagates the same constraint, returning HTTP 400 with
"this model is not supported MaxTokens, please use MaxCompletionTokens".
Lower the compatibility shim into `RemoteAPIChat.BuildChatCompletionRequest`
(the single OpenAI-protocol egress) so every internal caller keeps using
`MaxTokens` uniformly:
- Add `provider.IsOpenAIReasoningOrGPT5Model` with precise prefix matching
(covers `gpt-5*`, `o1`/`o1-*`, `o3`/`o3-*`, `o4`/`o4-*`; rejects
`olympus-1`, `openai-*`, `o3xtra`, etc.).
- When the provider is `openai`/`azure_openai` and the model matches, map
`MaxTokens` to `MaxCompletionTokens` (explicit `MaxCompletionTokens`
wins) and skip the unsupported sampling fields; `omitempty` keeps them
off the wire.
- Behavior is unchanged for gpt-4o, gpt-4, and all other providers.
Add unit tests for the matcher and for the request-building path
(Azure gpt-5.2, OpenAI gpt-5/o1-mini/o3/o4-mini, plus negative cases).
Fixes#1283
To diagnose where latency lives between "user hits send" and "first
token appears" we need a single number that can be matched across
the browser console and the server log. Add correlated TTFB markers
keyed by X-Request-ID:
Frontend (streame.ts)
* Generate the X-Request-ID once and reuse it for logging.
* Log request:start when fetchEventSource is invoked, response:headers
when onopen fires, and response:first_answer the first time an SSE
payload with response_type === 'answer' arrives. Filtering by event
type avoids treating session_title / references / tool_call as the
"first token".
Backend (session handler)
* parseQARequest records the wall-clock entry time and logs TTFB:start
with the same X-Request-ID.
* qaRequestContext carries receivedAt through to the stream handler.
* AgentStreamHandler emits a one-shot TTFB:first_answer_chunk log the
first time it observes an AgentFinalAnswerData chunk, so the delta
against TTFB:start is the server's request-in → first-token-out
budget.
The frontend delta minus the backend delta is then attributable to
network + gin middleware, which were previously invisible.
Previously the Langfuse timeline for a knowledge-chat request only
showed generations (chat.completion, embedding.embed, rerank), so the
work happening between them — query setup, vector/keyword search,
result merging, prompt assembly — appeared as unexplained gaps.
Add spans around the pieces that fill those gaps:
* qa.setup wraps request-time KB / model / search-target resolution
before the pipeline event loop starts, accounting for the visible
delay before the first chat.completion.
* pipeline.<event> wraps each pipeline stage's eventManager.Trigger
call so generations inside a stage nest under it. CHAT_COMPLETION_
STREAM is intentionally skipped because its OnEvent returns as soon
as the streaming goroutine starts; a stage span would always finish
before the chat.completion.stream generation it nominally parents.
* retrieve wraps the actual vector + keyword retrieve call inside
HybridSearch, exposing the DB round-trip that previously sat invisibly
between embedding generations and rerank.
* web_search wraps the external web-search HTTP call when enabled.
Quick-answer (RAG) agents can now configure a separate chat model for
the query-understanding stage (rewrite + intent classification),
decoupling it from the main conversation model so users can route the
lightweight rewrite call to a cheaper / faster model.
- Add CustomAgentConfig.QueryUnderstandModelID and plumb it through
PipelineRequest and ChatManage.Clone.
- query_understand plugin prefers QueryUnderstandModelID on the
text-only path; falls back to ChatModelID if the configured model
cannot be resolved (with a warn log). Multimodal path is unchanged
to keep vision-capable model selection intact.
- AgentEditorModal exposes a ModelSelector under the existing Query
Rewrite block; empty means reuse the main chat model.
- Add i18n strings (queryUnderstandModel / placeholder / desc) for
zh-CN, en-US, ko-KR, ru-RU.
- Extend Go SDK AgentConfig with the new field.
The data-analysis pipeline stage is a retrieval-strategy concern (it only
runs after chunk search/rerank), so the toggle belongs alongside the other
retrieval knobs rather than under knowledge base settings.
Refs: https://github.com/Tencent/WeKnora/issues/1244
The legacy in-pipeline DuckDB SQL data-analysis stage used to run on every
quick-answer RAG request whose retrieved chunks included a CSV/Excel file,
adding one extra LLM round-trip (~3s) to generate a SQL query that most
plain Q&A users never wanted. There was no way to disable it.
Introduce a per-agent DataAnalysisEnabled flag (default off), wire it
through PipelineRequest, and gate the DATA_ANALYSIS stage on it. Surface
the toggle in the agent editor for quick-answer agents with at least one
knowledge base attached.
Refs: https://github.com/Tencent/WeKnora/issues/1244
The request logger's sanitizeBody only matched lowercase / snake_case
field names (api_key, apikey, access_token, ...), so values for the
camelCase JSON fields actually used by the API (apiKey, secretKey,
refreshToken, accessToken, ...) were written to logs in clear text.
Replace the per-field patterns with a single case-insensitive regex
that tolerates optional `_`/`-` separators, covering snake_case,
camelCase and PascalCase variants, and extend coverage to id_token,
client_secret, private_key, auth_token, api_secret and passwd. The
field name is preserved; only its value is replaced with "***".
Add unit tests for sanitizeBody covering the previously-leaking
camelCase fields and common variants.
Fixes#1287
Add three new optional filters to the document list under a knowledge base
detail page — parse status, source/channel, and updated time range — and
rework multi-select to no longer cause the card title to jitter on hover.
Backend
- Introduce types.KnowledgeListFilter to aggregate optional filter dimensions
(tag, keyword, file_type, parse_status, source, updated_from/to) and switch
ListPagedKnowledgeByKnowledgeBaseID (repository/service/interface) to accept
it instead of a growing positional parameter list.
- The ListKnowledge HTTP handler accepts new parse_status, source, start_time
and end_time query params; time params accept RFC3339, "YYYY-MM-DD HH:MM:SS"
and "YYYY-MM-DD". The repository routes source="manual"/"url" onto the type
column to stay consistent with file_type semantics; other source values match
the channel column.
- Update the four other callers (agent_service, initialization) to pass an
empty filter struct, preserving prior behavior.
Frontend
- Add three controls in the doc-filter-bar (status select, source select,
date-range picker with future-date disabled) wired through getKnowled /
listKnowledgeFiles into the new backend params.
- Replace the hover-triggered card checkbox with an explicit "批量管理" mode
(mirrors the session list UX): in card view the checkbox only renders while
batch mode is on, entered via the per-card "..." menu; the list view keeps
its leading checkbox column. Switching from list to grid auto-enables batch
mode when something is already selected, so the selection stays visible.
- DocumentBatchBar now stays open whenever batch mode or selection > 0, and
its "取消选择" button both clears the selection and exits batch mode.
API surface sync
- Regenerate Swagger artifacts (docs/docs.go / swagger.json / swagger.yaml).
- Update docs/api/knowledge.md with the new query parameters.
- Add backward-compatible ListKnowledgeWithFilter + KnowledgeListFilter to the
Go SDK; the existing ListKnowledge keeps its signature.
i18n
- New filter labels in zh-CN / en-US / ko-KR / ru-RU; reuse existing
menu.batchManage / batchManage.cancel for the multi-select strings.
- Convert issue and PR templates to English only for broader reach
- Bug report: add Steps to Reproduce, Actual Behavior, WeKnora Version,
and Deployment Method as required fields; expand log guide to cover
Lite / Desktop / source builds
- Feature request: replace user-selected Priority with Impact to reduce
severity inflation
- PR template: slim down to 6 sections, add Conventional Commits hint
in the title, and require `make fmt && make lint && make test` in the
checklist
When image bytes for a multimodal task cannot be read via FileService
(e.g. tenant.StorageEngineConfig.MinIO is empty while the image was
saved using the global MINIO_* env vars), the previous code fell back
to the HTTP downloader, which rejected the provider:// URL with
"unsupported URL scheme". The asynq handler then returned an error,
asynq retried until exhaustion, and the per-knowledge "pending images"
counter was never decremented — leaving the document stuck in
"processing" indefinitely (issue #1282).
Changes:
- ImageMultimodalService now holds a default FileService and
resolveFileServiceForPayload falls back to it when the tenant-scoped
storage config cannot produce a usable service, mirroring the
write-side fallback in knowledgeService.resolveFileService.
- Extract readImageBytes: provider:// URLs are read exclusively via
FileService and never handed to the HTTP downloader.
- On unrecoverable read failure for a single image, log and skip that
image but still call checkAndFinalizeAllImages so the parent
knowledge can progress to post-processing.
Fixes#1282
Previously the custom agent editor hid the rerank model field when no
RAG-type knowledge base existed in the configured scope, but the
not-ready check and runtime hard-failed whenever knowledge_search was
in allowed_tools. Users with wiki-only or empty scopes saw "Rerank
Model required" warnings they could not resolve, and "All knowledge
bases" agents broke later if a RAG-type KB was added.
- Backend: when knowledge_search is enabled, fall back to the tenant
default rerank model (ConversationConfig.RerankModelID) before
erroring out, matching the built-in agent behaviour.
- Editor: always show the rerank field once a KB scope is selected;
only mark it required (red *) when the scope contains a RAG KB, with
a hint explaining the tenant-default fallback.
- Editor: only render rerank top_k / threshold sliders when a rerank
model is actually selected.
- Input field: drop the eager "missing rerank" not-ready reason; the
backend is now the single source of truth for rerank availability.
- i18n: add agent.editor.rerankModelOptionalHint across all locales.
`metadata->'similar_questions'::text` parses as
`metadata -> ('similar_questions'::text)` because `::` binds tighter
than `->`, so the expression yields jsonb instead of text. ILIKE on
jsonb then fails with "operator does not exist: jsonb ~~* unknown",
returning 1007 Internal server error for `search_field=similar_questions`
and `search_field=answers`. The `standard_question` branch was unaffected
because it uses `->>` (text) directly.
Wrap the json access in parens so the cast applies to the extracted
value: `(metadata->'similar_questions')::text ILIKE ?`. The MySQL
branch uses JSON_EXTRACT and is unchanged.
Fixes#1264
Quick-answer agent mode retrieves purely through vector/keyword chunk
search and ships with no `allowed_tools`, so the existing capability
filter (which only reads from `allowed_tools`) let wiki-only KBs through
in every entry point. End result: users could @-mention, select, and
receive suggested questions from wiki-only KBs in quick-answer mode,
but the underlying retrieval always returned empty.
Treat "RAG-only" as an implicit property of `agent_mode = quick-answer`
and union it with the tool-derived filter. The same predicate is now
used everywhere the user can pick or be steered toward a KB:
Backend
- `tools.DeriveKBFilterForAgent` / `KBSatisfiesAgentRequirements`
layer the implicit quick-answer requirement on top of tool derivation.
- `ListKnowledgeBases`, `SearchKnowledge` (shared-agent `@file`),
`resolveKnowledgeBasesFromAgent` (chat runtime), `/search` IM command,
and `GetSuggestedQuestions` now all use the agent-mode-aware variant.
- `GetSuggestedQuestions` also skips the wiki-page fallback for
quick-answer agents to cover the `selected` / explicit-kb-ids paths
where a wiki-only KB could still slip through.
Frontend
- `deriveKbFilterForAgent` / `kbSatisfiesAgentRequirements` mirror the
Go helpers.
- `@` mention dropdown (`Input-field.vue`) uses the new helper.
- Agent editor's "specified KB" picker (`AgentEditorModal.vue`) grays
out wiki-only KBs for quick-answer agents with a tooltip, and the
pre-save warning fires for quick-answer mode too.
- i18n: add `agentEditor.agentType.kbMismatch.quickAnswer` across all
four locales.
- Updated `pydantic` from 2.12.3 to 2.13.4 and `pypdfium2` from 5.0.0 to 5.8.0 in the `docreader` requirements.
- Modified the dependabot configuration to set `open-pull-requests-limit` to 0 and added an `ignore` rule for version updates across all ecosystems, allowing only security updates.
- Adjusted settings for `server-security`, `client-security`, `frontend-security`, and `miniprogram-security` groups to streamline security update handling.
This change aims to enhance dependency management and maintain security while reducing noise from version update PRs.
Agent mode does not persist `rendered_content` for user messages, so
when the next turn's history was rebuilt from DB, attachments uploaded
in prior turns disappeared — the model only saw the raw query plus the
prior assistant reply, breaking follow-up questions that referenced
the file (e.g. "what is in there?").
Reconstruct the attachment prompt from the stored `Attachments` column
when `RenderedContent` is empty, mirroring how image captions are
already replayed. KnowledgeQA turns (which do persist
`RenderedContent`) are unaffected and won't get attachments injected
twice.
Refs #1237
Three defects surfaced during end-to-end RAG verification — the first two
block real chat usage, the third makes Linux CI flaky:
1. KB id detection — `IsKBID` was checking
`strings.HasPrefix(s, "kb_")`, but WeKnora generates KB ids as bare
UUIDs (internal/types/knowledge_base.go: `uuid.New().String()` stored
in a `varchar(36)` column). Real ids therefore fell through to the
name-resolution path:
$ weknora chat ... --kb a32a63ff-fb36-4874-bcaa-30f48570a694
Error: knowledge base not found: a32a63ff-...
Switched the discriminator to a UUID regex
(`^[0-9a-fA-F]{8}-…-[0-9a-fA-F]{12}$`). KB names are arbitrary
user-supplied strings, so the canonical 8-4-4-4-12 form is an
unambiguous signal. Mirrors gcloud `--project`'s id-vs-name detection.
2. SSE terminal-frame — the accumulator's `Append` was gating
finalization on `r.Done`, but the server's KnowledgeQAStream protocol
emits a leading `agent_query` frame with `done=true` to deliver
session + message metadata *before* the answer fragments arrive:
event: message
data: {"response_type":"agent_query","content":"","done":true,…}
event: message
data: {"response_type":"answer","content":"你好","done":false}
…
event: message
data: {"response_type":"complete","content":"","done":true}
The accumulator therefore flipped to `finished=true` on frame #1 and
discarded every subsequent answer fragment — `weknora chat … --json`
returned `answer: ""` even though the LLM reported completion_tokens
> 0. Fixed: terminate only on `response_type == complete`.
References still captured opportunistically (they may arrive on a
dedicated `references` event before the terminator).
3. doctor credential_storage CI isolation — the check probes the real
OS keyring via `secrets.NewBestEffortStore()`: present on macOS dev
machines → StatusOK; absent on Linux CI runners without libsecret /
Gnome-Keyring → StatusWarn ("falling back to file store"). That
host-dependence was leaking into two test classes that assumed
StatusOK:
* cmd/doctor/doctor_test.go: TestDoctor_AllOK and
TestDoctor_NoConfig_StillRunsCredentialStorage already had a
withCredStoreFactory seam but didn't use it. Added the pin.
* acceptance/contract/envelope_test.go: doctor.success_offline
and doctor.error_network golden cases. The contract test runs
through the cobra tree in-process and shares cmd/doctor's
package-level credStoreFactory var — but couldn't reach it
because the existing seam was unexported.
Fix: export `doctor.SetCredStoreFactoryForTest(fn) (restore func())`
for out-of-package tests; acceptance/contract/helpers_test.go adds
a TestMain that pins the factory to a MemStore-returning closure
for the whole suite (MemStore is neither *FileStore nor a real
keyring, so doctor's type-switch hits StatusOK). Production stays
at secrets.NewBestEffortStore — only the test hook is now reachable
from across packages.
Test fixtures and goldens that used the old `kb_xxx` literals or
`Done: true` terminators were rewritten to use real UUIDs and
`ResponseType: ResponseTypeComplete` respectively. Per-command --help
text and Long descriptions / Examples now show a UUID rather than
`kb_…` so users see the correct shape from the start. New
TestAccumulator_IgnoresAgentQueryDone pins the SSE terminator bug so
it can't regress.
Tests: 24 cli packages green on macOS dev + Linux/macOS/Windows CI
matrix. Verified end-to-end against a live WeKnora server: `weknora
chat "..." --kb <UUID> --no-stream --json` returns the full LLM answer
in the envelope, live token streaming in TTY mode works, and the
credential_storage check renders deterministic envelopes across hosts.
Earlier per-dep dependabot PRs against the existing cli/, frontend/,
and miniprogram/ ecosystems flooded the review queue with 40+ PRs at
once. Apply a noise-bounding pattern uniformly to every ecosystem the
repo declares:
* **Monthly schedule** for routine version updates. Mirrors caddy /
hashicorp-terraform-actions: explicitly chosen over weekly to keep
maintainer review-attention bounded for a contributor-driven OSS
project. Among 10 surveyed mainstream configs (daily 7 / weekly 1
/ monthly 2), monthly is the canonical "minimum-noise" choice for
the contributor cadence WeKnora actually has.
* **Two `groups` per ecosystem** so per-dep PRs never flood:
- `<scope>-deps` bundles minor + patch updates → ONE PR
per ecosystem per month.
- `<scope>-deps-major` bundles semver-major bumps the same way.
Mirrors grafana's `*-breaking` pattern:
surfaces breaking changes as a single
review-required PR rather than silently
ignoring them. (gh-cli's `ignore:
semver-major` style means majors never
surface until someone manually edits
the .yml or the dep — easier to forget
than to act on. Grouping forces them
into the review queue.)
* `open-pull-requests-limit: 3` (was default 5) as a safety net so a
stalled review queue can't pile up stale group PRs.
* `commit-message.prefix: "chore(deps)"` so dependabot PR titles
match the repo's Conventional Commits style out of the box.
npm gets `prefix-development: "chore(deps-dev)"` for devDependency
bumps.
* Note: GitHub Security Advisory CVEs open immediately regardless of
`schedule.interval` — that field governs *version* updates only,
so the monthly cadence does NOT delay CVE response.
Coverage:
- gomod: /, /cli, /client
- npm: /frontend, /miniprogram
- pip: /docreader
- github-actions: /
Realistic steady-state with this config: 1–3 grouped PRs per month
total across all 7 ecosystems (vs. 40+ before), with major bumps
appearing as their own group PR every few months when upstream cuts
breaking releases.
Tests / build: not affected (config-only change). YAML validated with
`python3 -c "import yaml; yaml.safe_load(...)"`.
Discoverability gaps surfaced by the pre-PR review:
- New cli/README.md: install (build-from-source / pre-built once shipped)
+ 5-minute quickstart (auth login → kb list → link → doc upload →
chat) + multi-context walkthrough + JSON envelope shape + agent /
scripting integration overview + dev workflow. Points readers at
cli/AGENTS.md for the full operational contract.
- Top-level README.md: new "⌨️ Command-Line Interface" section between
Key Features and Getting Started, with a one-paragraph pitch + four
representative commands and links to cli/README.md and cli/AGENTS.md.
English README only this round; CN / JA / KO translations to follow
in v0.3 to match the existing four-language pattern.
- CHANGELOG.md [Unreleased] gets a "weknora CLI v0.2" bullet listing
the headline capabilities (10-command surface, project-link,
envelope, agent affordance, multi-context auth, doctor) and pointing
at cli/README.md.
- cli/AGENTS.md gains an "Architecture decisions" section documenting
ADR-3 (gh as primary mainstream north star + the four documented
deviations: link, chat/search, context use, doctor) and ADR-4
(Factory closures + narrow Service interfaces). The in-source
references (`(v0.2 ADR-3)`, `(per ADR-4)`) now point at committed
prose rather than dangling.
gh / lark / gcloud / stripe all ship a logout command and a way to
enumerate stored credentials on day one. WeKnora's `auth` subtree had
only login + status, leaving no documented purge path for keyring
secrets — a real concern for `--with-token` (sk-…) and JWT flows that
write credentials to OS keychains.
auth logout [--name <ctx>] [--all] [--json]
Clears keyring + file-fallback secrets (access / refresh / api_key
slots) for the named context (default: current) or every context
with --all. Removes the context entry from ~/.config/weknora/config.yaml
and clears current_context if the removed entry was active.
Mirrors `gh auth logout` and `lark auth logout`. As gh documents,
this does NOT revoke server-side — for API keys users must rotate in
the server UI, JWTs continue to be accepted until expiry.
auth list [--json]
Renders a compact table (NAME / HOST / USER / MODE) with the active
context marked `*`. Reads only config.yaml — no network, no keyring
touch. Mode is inferred from which credential ref is set (api_key
→ "api-key", token → "password"; both → "password" wins).
Mirrors gh's per-host enumeration (gh auth status iterates accounts)
and lark `auth list`. For weknora the contexts file already had this
data — the command is a thin renderer to match user muscle memory.
Deferred to a follow-up release:
- auth refresh + transparent 401 retry in the SDK (we already persist
refresh_token at login but never spend it; explicit gap)
- login --web browser OAuth flow (requires a server-side endpoint)
- auth token printer (cheap; defer with the rest)
Tests: 24 cli packages green. New: cmd/auth/logout_test.go (current
context, named, --all, no-contexts, unknown-name, no-current-no-flag,
mutex flags) + cmd/auth/list_test.go (human render, empty, JSON
envelope, inferMode edge cases). AGENTS.md command-surface note adds
the four-command auth subtree; screenshot section 4 adds `auth list`
alongside `auth status`.
Empirical mainstream-CLI surveys (gh / kubectl / aws / gcloud / stripe /
flyctl / terraform / vercel / netlify / lark) drove five alignment
fixes — each replaces a weknora-only design choice that mainstream CLIs
do not share. No backwards-compat shims; the CLI has no v0.1 users yet.
1. Single --kb flag (was --kb-id + --kb mutually exclusive)
Survey: 0/7 mainstream CLIs use two parallel flags for "by id" vs
"by name". Single flag (gh -R, gcloud --project) or positional
(kubectl, stripe, terraform). Closest analog — gcloud --project —
collapses identifier types onto one flag.
Now: every command exposes one --kb flag; client-side prefix
detection (cmdutil.IsKBID looks for "kb_") routes id-form values
through directly and name-form values through ListKnowledgeBases.
Mirrors gcloud --project's id-or-name auto-detection.
Touched: search, chat, doc list / upload / delete, link.
Factory.ResolveKB chain trimmed from 5 levels to 4.
2. link supersedes init
Survey: only vercel and netlify ship both `init` AND `link` as
siblings, and they keep them semantically distinct. weknora's pair
wrote the same .weknora/project.yaml file with the same meaning,
differentiated only by interactivity — that's a flag concern, not
a command concern.
Now: cmd/init/ deleted. cmd/link absorbs the interactive flow:
- link --kb <id-or-name> → non-interactive write
- link on a TTY → interactive prompt (lists KBs)
- link non-TTY without --kb → CodeKBIDRequired
Always overwrites silently (matches vercel link / netlify link /
kubectl apply rather than git init's refuse-if-exists).
Dead code purged: --force flag, CodeProjectAlreadyLinked error code.
3. whoami dropped
Survey: 7/7 mainstream CLIs ship exactly one identity command —
never both a status and a whoami. gh / gcloud / stripe pick status
(config + live API); aws / kubectl / flyctl pick whoami (live API).
weknora's auth status was already a superset of whoami (host +
context + user + email + tenant_id + tenant_name vs user_id +
tenant_id), so dropping whoami preserves all functionality and
aligns with the gh / gcloud / stripe form.
4. kb get alias dropped
`view` was already primary (gh repo view / gh pr view convention);
`get` was kept as a cobra alias for v0.0/v0.1 callers. With no
v0.0/v0.1 users to break, the alias is just noise on the command
surface. Acceptance contract envelope cases renamed kb_get.* →
kb_view.*; goldens renamed in lockstep.
5. api refactored to gh shape (-X/--method, default GET, auto-POST)
gh CLI's signature is `gh api <endpoint> [--method M]` — single
positional path, method as a flag, default GET, auto-promoted to
POST when a body is supplied. weknora's previous `api <method>
<path>` inverted this and forced the method to be passed even for
GET — a needless deviation from our declared north star.
Now: `api <path> [-X METHOD] [--data ...]`. Exit-10 protocol
on the DELETE escape-hatch is preserved; -X DELETE still hits
ConfirmDestructive when -y absent.
Plus: AGENTS.md gains an explicit note that `doctor` is a deliberate
divergence from gh / lark — borrowed from `flutter doctor` / `brew
doctor` because RAG deployments routinely break on misconfigured
embeddings / storage / credentials and a 4-status structured envelope
is the cleanest surface for it.
Tests: 24 cli packages green (was 26 in PR-14; init + whoami packages
removed). Acceptance contract envelope cases for whoami removed,
kb_get → kb_view renamed, search args / mock path updated for the
kb_<id> form. e2e harness flag args updated. Factory.ResolveKB tests
rewritten for the single-flag shape. api_test driver updated for the
positional-path / -X-method shape.
Foundation PR-1 reserved several internal packages and helpers as
scaffolding for follow-up PRs that ended up taking different routes.
Audit confirms zero production references; this commit removes them so
the cli/ tree reflects what's actually shipped.
Removed (148 LOC):
cli/internal/safepaths/ — `Validate` / `WithinRoot` /
three sentinel errors. Reserved
for `weknora doc upload`'s path
scrubbing; that command landed
in PR-10 using its own
`validateUploadPath` (os.Stat +
regular-file check) — sufficient
for the actual threat model
(local CLI invocations).
cli/internal/cmdutil/json_flags.go — `AddJSONFlags` helper +
unused --jq / --template flag
registration. Reserved for PR-3
"lipgloss tables / jq evaluator"
which never materialized; every
command directly registers
BoolVar(&JSONOut, "json", ...)
since v0.0 ship time.
cmdutil.NewTableExporter — empty alias for jsonExporter,
reserved for the same PR-3
renderer. Removed; jsonExporter
stays under NewJSONExporter.
cmdutil.Options marker interface — empty interface{} reserved as a
convention; no command embeds
or asserts against it.
Stale comments fixed:
- cmd/root.go: package comment updated kb (list+get) → kb
(list+view+create+delete) and noted the `get` cobra alias.
- cmd/root.go: dropped --no-version-check forward-reference (no such flag).
- cmd/root.go: removed "(PR-7)" attribution from NewRootCmd doc comment.
- cmd/kb/kb.go: same package-comment update.
- cmd/chat/chat.go: replaced "PR-7" mention in --help example with a
generic placeholder so cobra-rendered help is review-clean.
- cmd/search/search.go: removed "Lipgloss tables arrive in PR-3"
forward-reference; the inline indent helper is the shipped form.
- internal/agent/annotations.go: ShouldUseAgentMode → DetectAIAgent
(removed in PR-12).
AGENTS.md "Known limitations" section added:
Documents that chat / search / doc upload currently surface server-side
precondition misses (LLM / vector store / storage engine not configured)
as `network.error` with `context deadline exceeded`. A planned future
release will introduce a `precondition.*` typed error namespace
(server returns HTTP 412 before opening the SSE / streaming response).
This documents the limitation honestly for reviewers and integrators
rather than claiming a behavior we don't yet have.
Tests: 27 cli packages pass (safepaths_test was the 28th — gone with the
package). go vet clean.
Borrows the lark-cli agent-affordance model
(https://github.com/larksuite/cli/blob/main/AGENTS.md +
skills/lark-shared/SKILL.md) so weknora is designed to be agent-friendly:
error messages, output format, and flag design follow conventions agents
can rely on.
cli/AGENTS.md (operational reference for LLM agents invoking weknora):
Public document covering envelope schema, exit-code protocol
(0/1/2/10/130), stdout/stderr separation, and behavioral rules.
Sensitive commands (\`context use\`, \`kb delete\`, \`doc delete\`, \`init\`)
gain "AI agents:" paragraphs in their cobra Long descriptions so
guidance shows in --help.
format.Envelope schema additions:
Risk per-operation classification (read / write / high-risk-write +
action description), populated by write commands on both success
and failure paths.
Notice system advisories (CLI update available, server-CLI version
skew); type defined, emit sites land in v0.3.
DryRun marker for envelopes returned from --dry-run preview paths.
RiskLevel constants realigned to lark's taxonomy: read / write /
high-risk-write (was: read / mutating / destructive — not yet wired by
any command).
cmdutil.Error gains OperationRisk; PrintErrorEnvelope auto-attaches it
to envelope.Risk so destructive failure paths surface uniformly.
Exit-10 confirmation protocol:
New ErrorCode \`input.confirmation_required\` mapped to exit code 10 in
cmdutil.ExitCode. ConfirmDestructive now returns this code (with
OperationRisk attached) when stdout is non-TTY or --json was set, with
-y/--yes absent. Previous behavior — silent proceed in non-TTY — was
unsafe: scripts and agents could delete resources with no explicit
approval. Three test cases re-pinned around the new contract.
This is a wire-contract change for any caller who relied on silent
proceed; v0.0/v0.1 had no destructive commands, so the blast radius is
contained to v0.2 itself.
--dry-run global flag:
cmd write paths (kb create/delete, doc upload/delete, api POST/PUT/PATCH/
DELETE) check cmdutil.IsDryRun(cmd) and skip the SDK call, emitting an
envelope with dry_run=true plus a Risk classification. Read commands
ignore --dry-run by design (no side effect to preview). Human-mode
prints \`[dry-run] would <action>\` to stdout.
Command discovery: agents introspect via the existing \`--help\` surface
(consistent with gh / kubectl / aws / gcloud / terraform — none of them
ship a CLI-tree self-description command). An earlier draft added a
\`weknora schema\` reflection command; dropped after a mainstream survey
found it has no stable analog (lark-cli's schema describes Lark API
methods, not its own CLI tree).
Tests: 27 cli packages pass at this commit. Added two new tests covering
envelope.risk and envelope._notice serialization.
Audited the v0.0~v0.2 21-command surface against gh / kubectl / cargo /
npm / git / docker / flyctl / vercel / supabase / brew. WeKnora was
cherry-picking from multiple heritages, producing an inconsistent feel:
the kb subtree mixed gh verbs (create / delete / list) with a kubectl
verb (get); confirmation flag duplicated --force (docker/kubectl) with
global -y/--yes (gh/vercel/npm); the --agent flag stretched Stripe's
telemetry-tag pattern into a behavior-mode switch that no mainstream
CLI does.
ADR-3 picks gh as the primary north star. Documented deviations remain
for project-link (vercel/cargo), chat (openai-cli), context (kubectl-
light), and doctor (brew/flutter). The decision and its deviations are
documented self-contained in cli/AGENTS.md.
Surface changes:
- kb get → kb view (gh repo view convention); "get" kept as cobra Alias
for v0.0/v0.1 callers — see https://cli.github.com/manual/gh_repo_view.
- kb delete --force / doc delete --force removed in favor of the global
-y/--yes persistent flag (gh repo delete --yes convention). One
mechanism skips destructive prompts; ConfirmDestructive's parameter
renamed `force` → `yes` to match.
- --agent omnibus mode-switch removed. Stripe's DetectAIAgent (the
cited inspiration) only tags User-Agent for telemetry, never flips
behavior; gh / kubectl / aws / docker / flyctl all decline this kind
of flag. The 7-env auto-detect list is reduced to the two entries
Stripe also recognizes (CLAUDECODE, CURSOR_AGENT) — the other five
had no agent-documented source. ApplyAgentSugar / ShouldUseAgentMode
and the dead --no-interactive / --no-progress globals are deleted
entirely.
- DetectAIAgent and SetAgentHelp annotations are kept: env detection
now only triggers AGENT-targeted help text rendering (no behavior
change), matching Stripe's narrower scope.
Tests: 27 cli packages green (acceptance/contract still pins kb get
golden; the alias keeps it valid).
Close the v0.2 RAG demo loop and ship the validation infrastructure:
- weknora api <method> <path> [--data X | --data-file F]
Raw passthrough wrapping client.Raw, gh-style. JSON envelope mode
surfaces status / headers / parsed body. Non-2xx routes through
cmdutil.ClassifyHTTPStatus (factored out of ClassifyHTTPError so
both SDK-error and direct-status paths stay aligned — reuse review).
- weknora chat <text> [--session-id S] [--no-stream]
KnowledgeQAStream consumer with two output modes:
- TTY default: token streaming + references footer
- --json / --no-stream / non-TTY: buffered single envelope
Auto-creates a session when --session-id is omitted; the id prints
to stderr at start AND on stream failure (^C scrolls past the
first announcement, so the recovery hint is re-surfaced when the
user is most likely to need it).
- cli/internal/sse/Accumulator
buffers Content / References / SessionID across SDK callbacks.
Idempotent post-Done so misbehaving servers don't corrupt state.
- doctor: ok → ok / warn / fail / skip
warn marks soft issues that don't block: server within compat range
but >=1 minor behind CLI; credential storage falling back to file
because keyring is unavailable. Envelope.ok stays true on warn,
flips false on fail (exit 1). doctor.error_network golden updated.
- cli/acceptance/e2e/ real-server RAG full loop
Build-tagged //go:build acceptance_e2e — kept out of the default
`go test ./...`. Exercises kb create → doc upload → poll ready →
search → chat → cleanup against a server pointed at by
WEKNORA_E2E_HOST / _TOKEN.
- .github/workflows/cli-e2e.yml
manual workflow_dispatch + label-gated PR trigger
("acceptance-e2e"). No-ops gracefully when the secrets aren't set
so cross-fork PRs can't accidentally fail the suite.
Add the resource-management surface to the v0.2 CLI:
- weknora kb create --name X [--description Y] [--embedding-model Z]
- weknora kb delete <id> [--force]
- weknora doc list [--kb-id X | --kb NAME] [--page N] [--page-size M]
- weknora doc upload <file> [--kb-id X | --kb NAME] [--name custom]
- weknora doc delete <id> [--force]
doc/* uses Factory.ResolveKB so the cwd's project link is honored when
--kb-id is omitted. doc upload validates path existence with os.Stat
(rejects directories; follows symlinks to mirror SDK os.Open behavior).
doc list sorts by updated_at desc so newer items surface first.
Both delete commands route through cmdutil.ConfirmDestructive — the
"destructive op needs explicit user opt-in" pattern was about to be
copy-pasted across the new subtree, so it was extracted with the delete
commands as their first consumer. Saves the same dedup pass when v0.3
adds session/agent delete. (PR-12 later renames the flag from --force
to the global -y/--yes for gh-style consistency.)
iostreams.SetForTestWithTTY pairs with the existing SetForTest helper:
the latter never reports stdout as a TTY (singleton replacement uses an
in-memory buffer), so the confirm-yes / confirm-no test branches need a
TTY-on variant.
v0.2 grounds the CLI's resource commands (kb / doc / chat / query) in a
per-project link file (.weknora/project.yaml) so users don't have to pass
--kb-id on every invocation. Mirrors npm/cargo/git: walk up the cwd tree
to find the project root, override via flag or env when needed.
This commit ships the foundation layer:
- cli/internal/projectlink/ Discover (walk-up, depth=64) / Load / Save
- cmdutil.Factory.ResolveKB 5-level fallback chain:
--kb-id flag → --kb name (ListKnowledgeBases lookup) →
WEKNORA_KB_ID env → walk-up project link → CodeKBIDRequired
- cmdutil.ResolveKBNameToID shared name→id helper used by init / link
/ Factory.ResolveKB (was duplicated 3 ways
in early implementation; reuse review #2)
- cli/cmd/init/ interactive (huh prompt) or flag-driven
first-time setup; refuses to overwrite
without --force
- cli/cmd/link/ non-interactive update; --kb-id and --kb
are mutually exclusive and one is required
Also registers the v0.2 ErrorCode set (all codes for the eight new
commands) and AST-scan identToErrorCode mapping in one place — keeps the
acceptance/contract suite green across the v0.2 commit chain even before
later commits reference each code.
The client SDK emitted 17 fmt.Printf / fmt.Println trace events directly to
stdout from KnowledgeQAStream and SearchKnowledge. Any consumer of the SDK —
the new CLI in particular — saw these lines interleaved with its own
output, breaking JSON envelope contracts and polluting human-mode tty.
Replace with a package-level slog logger. Default writes to io.Discard so
existing callers see no output change; set WEKNORA_SDK_DEBUG=1 to route
events to stderr for development.
No public SetLogger() API: the trace events were dev-only and a setter
would expand the SDK surface for a one-off use case.
The Agent ran a parallel context cache (Redis/in-memory) on top of the
messages table to feed multi-turn history into the LLM. That dual-write
caused subtle drift (e.g. compression diverging from DB, system-prompt
swaps lost on restart) and required a separate ClearContext path on
session/IM clear.
Make the messages table the single source of truth:
- Add service.LoadAgentHistory: rebuilds chat.Message history per turn
from the persisted messages, expanding AgentSteps into proper OpenAI
assistant_with_tool_calls + tool messages and replaying the canonical
final answer (with <think> blocks stripped). final_answer tool calls
are filtered to avoid duplicating the trailing answer.
- Make AgentEngine stateless across turns: drop ContextManager / sessionID
cache plumbing from the engine, agent service, and CreateAgentEngine
signature. The engine only uses sessionID for logging/event emission.
- Wire AgentQA to load history from DB on demand using HistoryTurns
(default 5) when MultiTurnEnabled, otherwise run with empty history.
- Delete the llmcontext package (ContextManager interface, Redis/memory
storage, factory) and the SessionService.ClearContext API path; IM
/clear and session message clear no longer need to invalidate cache.
Behavior preserved: KnowledgeQA-mode replay is unchanged (turns with
empty AgentSteps just produce the single canonical assistant message),
and Agent-mode turns now consistently see prior tool calls and results.
Mainland China cloud VMs (Tencent Lighthouse, Aliyun, etc.) frequently
cannot reach get.docker.com, github.com, or even community GitHub
mirrors like gh-proxy.com. The cloud-image bootstrap previously had no
escape hatch for this and failed at the very first curl.
This adds a new DOCKER_INSTALL_MIRROR env var to prepare.sh. When set,
it skips get.docker.com and installs docker-ce + compose-plugin from an
apt mirror of Docker's official repo (e.g. mirrors.tencent.com,
mirrors.aliyun.com).
README.md also gets:
- A GH_PROXY env var threaded through bootstrap methods A and B so the
initial script pull can route through gh-proxy / ghfast.
- An explicit recommendation to prefer method C (scp from local) on
mainland China VMs.
- A consolidated "三件套" table mapping WEKNORA_GH_PROXY /
DOCKER_INSTALL_MIRROR / DOCKER_REGISTRY_MIRROR to per-cloud
endpoints, so users hit one place to copy the full env.
`${SEARXNG_SECRET:?...}` made the variable mandatory at compose parse time,
which forced *any* compose command (default profile included) to fail when
SEARXNG_SECRET was unset, with a message confusingly claiming the searxng
profile was being started.
Switch to `${SEARXNG_SECRET:-weknora-default-searxng-secret-...}` so the
searxng profile starts zero-config. Default deployments bind searxng to
127.0.0.1 only, so a shared default secret is acceptable; .env.example
now explicitly warns to rotate it before flipping SEARXNG_BIND=0.0.0.0,
since secret_key signs image-proxy URLs.
The SearXNG provider forwarded `parameters.api_key` as
`Authorization: Bearer <key>` for reverse-proxy auth, but the frontend
never rendered the API key input for self-hosted providers
(`requires_api_key=false`), so the path was unreachable from the UI.
Rather than expose another UI knob for a niche reverse-proxy setup that
the project does not currently support, remove the Authorization header
branch and the `apiKey` field on `SearxngProvider`. Tenants who put
SearXNG behind an authenticating reverse proxy can still front it with
network-level auth (mTLS, IP allowlist) without changes here.
`WebSearchProviderParameters.APIKey` is generic and used by other
providers, so it stays.
- Adjusted the formatting of struct fields in the SSRF whitelist test cases for improved readability and consistency.
- No functional changes were made; this is purely a code style improvement to enhance maintainability.