3 Commits

Author SHA1 Message Date
JackZhao10086 a6de8360f0 feat(auth): add scope hint for missing authorization errors (#776)
* feat(auth): add scope hint for missing authorization errors

* fix(auth): handle existing hints in missing scope error

* refactor(auth): centralize user authorization error detection

* fix(auth): handle nil error case in IsNeedUserAuthorizationError
2026-05-08 15:23:29 +08:00
liuxinyanglxy 4d4508dfd7 feat(event): add event subscription & consume system (#654)
* feat(event): add event subscription & consume system with orphan bus detection

Introduces end-to-end Feishu event consumption via a new `lark-cli event`
command family. Users can subscribe to and consume real-time events
(IM messages, chat/member lifecycle, reactions, ...) in a forked bus
daemon architecture with orphan detection, reflected + overrideable JSON
schemas, and AI-friendly `--json` / `--jq` output.

Commands
--------
- `event list [--json]`      list subscribable EventKeys
- `event schema <key>`       Parameters + Output Schema + auth info
- `event consume <key>`      foreground blocking consume; SIGINT/SIGTERM
                             /stdin-EOF shutdown; `--max-events` /
                             `--timeout` bounded; `--jq` projection;
                             `--output-dir` spool; `--param` KV inputs
- `event status [--fail-on-orphan] [--json]`   bus daemon health
- `event stop [--all] [--force] [--json]`      stop bus daemon(s)
- `event _bus` (hidden)      forked daemon entrypoint

Architecture
------------
- Bus daemon (internal/event/bus): per-AppID forked process that holds
  the Feishu long-poll connection and fans events out to 1..N local
  consumers over an IPC socket. Drop-oldest backpressure, TOCTOU-safe
  cleanup via AcquireCleanupLock, idle-timeout self-shutdown, graceful
  SIGTERM.
- Consume client (internal/event/consume): fork+dial the daemon,
  handshake, remote preflight (HTTP /open-apis/event/v1/connection),
  JQ projection, sequence-gap detection, health probe. Bounded
  execution (`--max-events` / `--timeout`) for AI/script usage.
- Wire protocol (internal/event/protocol): newline-delimited JSON
  frames with 1 MB size cap and 5 s write deadlines. Hello / HelloAck /
  PreShutdownCheck / Shutdown / StatusQuery control messages.
- Orphan detection (internal/event/busdiscover): OS process-table scan
  (ps on Unix, PowerShell on Windows) with two-gate cmdline filter
  (lark-cli + event _bus) that naturally rejects pid-reused unrelated
  processes.
- Transport (internal/event/transport): Unix socket on darwin/linux,
  Windows named pipe on windows.
- Schema system (internal/event, internal/event/schemas): SchemaDef with
  mutually-exclusive Native (framework wraps V2 envelope) or Custom
  (zero-touch) specs. Reflection reads `desc` / `enum` / `kind` struct
  tags, with array elements diving into `items`. FieldOverrides overlay
  engine addresses paths via JSON Pointer (including `/*` array
  wildcard) and runs post-reflect, post-envelope. Lint guards orphan
  override paths.
- IM events (events/im): 11 keys — receive / read / recalled, chat and
  member lifecycle, reactions — all with per-field open_id / union_id /
  user_id / chat_id / message_id / timestamp_ms format annotations.

Robustness
----------
- Bus idle-timer race fix: re-check live conn count under lock before
  honoring the tick; Stop+drain before Reset per timer contract.
- Protocol frame cap: replace `br.ReadBytes('\n')` with `ReadFrame` that
  rejects frames > MaxFrameBytes (1 MB). Closes a DoS path where any
  local peer could grow the reader's buffer unbounded.
- Control-message writes gated by WriteTimeout (5 s) so a wedged peer
  kernel buffer can't stall writers indefinitely.
- Consume signal goroutine: `signal.Stop` + `ctx.Done` select, no leak
  across repeated invocations in the same process.
- JQ pre-flight compile so bad expressions fail before the bus fork and
  any server-side PreConsume side effects.
- `f.NewAPIClient`'s `*core.ConfigError` now passes through unwrapped
  so the actionable "run lark-cli config init" hint reaches the user.

Subprocess / AI contract
------------------------
- `event consume` emits `[event] ready event_key=<key>` on stderr once
  the bus handshake completes and events will flow. Parent processes
  block-read stderr until this line before reading stdout — no `sleep`
  fallback needed.
- All list-like commands have `--json` for structured consumption.
- Skill docs in `skills/lark-event/` (SKILL.md + references/) brief AI
  agents on the command surface, JQ against Output Schema, bounded
  execution, and subprocess lifecycle.

Testing
-------
Unit tests across bus/hub, consume loop, protocol codec, dedup,
registry, transport (Unix + Windows), schema reflection, field
overrides, pointer resolver. Integration tests cover fork startup,
shutdown, orphan detection, probe, stdin EOF, preflight, bounded
execution, and Windows busdiscover PowerShell compatibility.

Change-Id: Ib69d6d8409b33b99790081e273d4b5b01b7dbf80

* fix(event): address CodeRabbit findings + lift patch coverage above 60%

CodeRabbit comments (PR #654)
-----------------------------
1. bus/dedup: IsDuplicate dropped legitimate (post-TTL) events after
   cleanupExpired fired. The run-every-1000-inserts cleanup removed
   TTL-expired IDs from the `seen` map but left them in the ring;
   IsDuplicate's ring-scan fallback then rediscovered them and falsely
   reported "duplicate", and bus.Publish silently dropped the event.
   Removed the ring-scan branch — `seen` is the sole authority, the ring
   only bounds map size via overflow eviction. New regression test
   TestDedupFilter_TTLExpiryAfterCleanupRunRespected exercises the 10-
   insert + cleanup path and guards the fix.

2. consume/remote_preflight: the decoder only read `data.online_instance_
   cnt`. A non-zero business code with no data payload decoded to 0 and
   callers treated it as "verified zero", forking a local bus that would
   duplicate events. Added Code / Msg fields and promoted code != 0 into
   an error so the caller distinguishes verified-zero from check-failed.

3. cmd/event/stop: swapped os.ReadDir / os.Stat to vfs.ReadDir / vfs.Stat
   in discoverAppIDs per project guideline (enables test mocking). New
   TestDiscoverAppIDs_* lifts discoverAppIDs from 0% to 100%.

4. cmd/event/appmeta_err: narrowed authURLPattern from
   feishu.cn|feishu.net|larksuite.com|larkoffice.com to the two hosts
   consoleScopeGrantURL actually produces. Kept the allowlist pinned to
   ResolveEndpoints' output with a comment flagging the synchrony.

5. cmd/event/list: moved "No EventKeys registered." and "Use 'event
   schema <key>' for details." hints to stderr so `event list | jq`
   style pipelines don't ingest them as data.

6. cmd/event/schema: runSchema is a RunE entry point; swapped the bare
   fmt.Errorf on resolveSchemaJSON failure to output.Errorf so AI
   agents parse a structured error envelope.

Coverage bumps (patch ~50% -> ~60%)
-----------------------------------
internal/event/consume/loop_test.go: loop.go was 0% at patch time.
New tests cover consumeLoop end-to-end via net.Pipe (events -> sink,
max-events -> ctx.Done -> PreShutdownCheck/Ack), seq-gap warning,
jq filtering + early compile failure, isTerminalSinkError classifier.
Takes consumeLoop from 0% to ~74%.

internal/event/protocol/messages_test.go: all NewXxx constructors,
Encode/Decode roundtrip per message type, EncodeWithDeadline deadline
enforcement, ReadFrame MaxFrameBytes rejection + EOF propagation.
Takes protocol from 28% to ~86%.

Also bundles small UX polish:
- cmd/event/consume: --output-dir flag doc flags path-traversal behavior;
  jq-validation failures now re-wrap with an event-specific hint
  pointing at `event schema` for payload shape.
- internal/event/consume.validateParams: error now names the EventKey
  and lists valid param names inline so AI callers recover without a
  second `event schema` round-trip.
- skills/lark-event: description expanded to mention
  listener/subscribe/consume synonyms + the IM scope set explicitly;
  lark-event-im reference polished; obsolete lark-event-subscribe
  reference removed.

Verified with go test -race -timeout 120s across ./cmd/event/...,
./events/..., ./internal/event/...; gofmt clean; go vet clean.

Change-Id: I3837b8645ea1d7529c9a8fd4c2bbfa965ae1b519

* test(event): cover format helpers + cobra factories

Adds cmd/event/format_helpers_test.go covering the pure output helpers
and factory wire-ups that RunE-level tests would need a live bus to
exercise:

- writeStopJSON: shape assertions + nil → [] (scripts expecting
  .results | length must not see null).
- writeStopText: stdout vs stderr routing — stopped / no-bus lines to
  stdout, refused / errored lines to stderr.
- busState.String: all three discriminator values.
- humanizeDuration: each bucket boundary (seconds / minutes / hours / days).
- writeStatusText: covers stateNotRunning / stateRunning (with consumer
  table) / stateOrphan (with kill hint).
- writeStatusJSON: orphan entry carries suggested_action + issue;
  running entry must NOT carry those fields (hint-leak guard for
  scripts that key on issue != "").
- exitForOrphan: flag-off never errors; flag-on errors iff any orphan
  is present, with ExitValidation code.
- NewCmdConsume / NewCmdStatus / NewCmdStop / NewCmdList / NewCmdBus:
  flag registration + RunE presence, so review catches flag-name drift.
  NewCmdBus check also pins Hidden=true.

Lifts cmd/event coverage 51.7% → 61.1%; aggregate event-package
coverage crosses the 60% codecov patch threshold (62% locally).

Change-Id: I9ecf3d905a8f9607b9441ee8a61e746496e2be63

* fix(event): address lint + deadcode CI failures

4 golangci-lint findings + 1 deadcode finding flagged on PR #654.

lint
----
1. cmd/event/stop.go:86 (ineffassign): `targets := []string{}` is
   overwritten by both branches of the `if o.all` below, so the empty-
   slice initializer is dead. Switched to `var targets []string`.
2. cmd/event/consume.go nilerr: the user-identity scope preflight
   swallows a non-nil ResolveToken error and returns nil. This is
   intentional — a missing/expired user token must not block consume;
   the bus handshake will surface the real auth error with actionable
   hints. Added `//nolint:nilerr` with a 4-line comment pinning the
   reasoning.
3. events/im/message_receive.go:62 nilerr: malformed JSON payload
   returns the original bytes + nil so consumers still see the event
   (the WARN breadcrumb lives in the outer loop). Added
   `//nolint:nilerr` with a one-line comment.
4. internal/event/schemas/fromtype_test.go:26 unused: `unexportedStr`
   is a reflection-test fixture — its presence (not value) exercises
   the FromType skip-unexported path verified at the "unexported
   field should not be in schema" assertion. Added `//nolint:unused`
   and a 4-line comment pointing at the guarded assertion.

deadcode
--------
5. internal/event/testutil/testutil.go: NewTCPFake has no callers in
   the repo. Removed the constructor plus the `inner == nil` TCP-mode
   branches from Listen / Dial / Cleanup. FakeTransport now only
   supports the wrapped-overlay mode (NewWrappedFake), which is the
   one every existing test uses. Doc comment simplified accordingly.

Verified locally: go test -race -timeout 120s across ./cmd/event/...,
./events/..., ./internal/event/... all green; gofmt clean; go vet
clean.

Change-Id: Ie8a2270827a0bde6b8159ab70aaf5c1e9ca7d5b9

* fix(event): drop stale enum + simplify protocol test type helper

- events/im/message_receive.go: dropped the `enum` tag on
  ImMessageReceiveOutput.MessageType. convertlib registers many more
  message types than the old 11-item list (video / location /
  calendar / todo / vote / hongbao / merge_forward / folder / ...),
  so a partial enum would tell AI consumers that valid values like
  "video" are invalid and produce false-negative JQ filters.

- internal/event/protocol/messages_test.go: collapsed the
  typeOf → reflectTypeName → stringType chain in
  TestEncode_DecodeRoundtripAllTypes to a single fmt.Sprintf("%T", v).
  The hand-maintained type switch silently returned "<unknown>" for
  any new message type, which would have let future Decode bugs slip
  past the roundtrip assertion. Also removed a dead `cases` table at
  the top of TestConstructors_PinTypeField left over from an earlier
  refactor.

Change-Id: I831e96f8417e80637596030d652a559de0d33122

* docs(event): polish skill docs + rename root_path_hint to jq_root_path

- skills/lark-event/SKILL.md, lark-event-im.md: translated to English,
  reorganized around a top-level "Core commands" table, scenario
  recipes tightened.
- cmd/event/schema.go: renamed the writeSchemaJSON hint field
  RootPathHint / "root_path_hint" -> JQRootPath / "jq_root_path" to
  make its purpose (a jq path prefix) obvious at the call site; no
  external consumer depends on the old name yet.

Change-Id: I00c14061ca33caedc0975bfeadc4b26d3dcd314d

* chore(event): strip excessive comments

Change-Id: I8f44f36f5dbdba3ef95dfc67069dc796232f91ec

* fix(event): dedup self-eviction race + protocol oversized-frame test

dedup: in IsDuplicate, the ring-slot eviction step deleted seen[id] even
when ring[pos] equalled the freshly-recorded id (post-TTL reinsertion
landing on its own historical slot). Net result: ring still held id but
seen did not, so the next IsDuplicate(id) returned false and the
duplicate was delivered. Skip the delete when old == eventID. New
TestDedupFilter_SelfEvictionPreservesFreshEntry pins the invariant by
pre-loading the ring slot and asserting the second call still reports
duplicate.

protocol: TestReadFrame_RejectsOversized used strings.Contains feeding
t.Logf, so any non-nil error passed — including a future regression
that returned io.ErrUnexpectedEOF while silently keeping the buffer
unbounded. Promoted MaxFrameBytes overflow to a sentinel
ErrFrameTooLarge and the test now asserts via errors.Is.

Change-Id: I50281dad392152b0ca083fd30c38eb0695e63bd3

* docs(event): clarify .content shape per message_type + add sender filter recipe

Change-Id: I619fd15c1a362e42e6602fd3e3316bbc75eddc5e

* fix(event): replace cmdline-regex bus discovery with PID file + close concurrent fork race

Bus discovery previously walked the OS process table and parsed `--profile cli_*` from
cmdline; the regex rejected any non-cli_ profile name (D-03a). Replace with per-AppID
bus.pid + bus.alive.lock under events/<AppID>/, probed via try-lock. AppID round-trips
through the directory name, so the profile-vs-AppID confusion is gone by construction.

Also fix B-07 (two consumers each fork an independent bus, halving event delivery):
- forkBus holds bus.fork.lock until child is dial-able, not just until cmd.Start
- bus daemon takes alive.lock before binding the socket; cleanup-TOCTOU race can no
  longer leave two listeners on different inodes

status.go renders an orphan with PID=0 distinctly (live bus but pid file unreadable)
so we never print "Action: kill 0".

Change-Id: I3bf0a6cf1d91fb274ac5a6df83d66896aafb291f

* style(event): gofmt bus.go

Trailing blank line introduced when appending acquireAliveLock helper.

Change-Id: I4ae1b4a4363dc6c89dcbd6a170f4563117490ba3

* fix(event): swap os.Remove/Rename for vfs.* and silence forbidigo on internal diagnostics

golangci-lint forbidigo blocks os.* in internal/. Switch the pid-file write to vfs.Remove/vfs.Rename and add a nolint marker on the two stderr diagnostics in busdiscover, matching the existing pattern in consume/*.

Change-Id: Ia6768be62aefeb8ca40f991d3130a78ef2ec0ea5

* fix(event): cross-platform --all + clean SIGPIPE shutdown for consume

- stop --all: replace bus.sock-file probe with busdiscover lock-based
  scan; previously skipped Windows entirely (named-pipe transport, no
  socket on disk) and misidentified Unix stale sockets as live. Same
  win for `event status` (shares discoverAppIDs).

- consume: ignore SIGPIPE so a closed stdout pipe (e.g. `... | head -n 1`)
  surfaces as EPIPE error and reaches the existing isTerminalSinkError
  cleanup path (log "output pipe closed", lastForKey query, hub
  unregister), instead of being killed by Go's default fd 1/2 SIGPIPE
  handler with exit 141 and zero deferred cleanup.
  Build-tagged: real on unix, no-op on windows (no SIGPIPE there).

Change-Id: I453b19f05c489fd9d5c1a9ba3bdc35e127c15b83

* docs(event): translate IM EventKey descriptions and field tags to English

Aligns with the rest of the codebase (titles, struct names, README) which
are already in English. Surfaces in `event list` / `event schema` and is
also consumed by AI agents.

- events/im/message_receive.go: 11 desc tags on ImMessageReceiveOutput
- events/im/native.go: 10 description fields on Native EventKeys
- events/im/register.go: im.message.receive_v1 Description

Change-Id: I6f46950b4793f137e0129c1f06019a3419195443

* docs(event): drop misleading AuthTypes[0] auto-default claim

The KeyDefinition comment and SKILL.md flag table both stated that
`--as auto` resolves to `AuthTypes[0]`. It does not — ResolveAs goes
through global rules (config default_as / credential hint / `bot`
fallback) without consulting the EventKey. AuthTypes is only used by
CheckIdentity as a post-resolve whitelist.

Reword the field comment to plain whitelist semantics and have SKILL.md
defer `--as` documentation to lark-shared.

Change-Id: Ia5d3d3790aed05813a0fa72d6b43518224e2055b

* revert(comments): restore original comments on 3rd-party files

e61482a stripped comments across 105 files. Restore the four files
authored by others (cmd/build.go, shortcuts/common/{types,runner}.go,
shortcuts/event/subscribe.go) to their pre-strip state so unrelated
documentation isn't churned in this PR.

Change-Id: Ie2527b06bfaf5b3861b0b9dff1e19bbfe7dde456
2026-04-28 11:19:02 +08:00
tuxedomm fbed6beac3 refactor: split Execute into Build + Execute with explicit IO and keychain injection (#371)
* refactor(cmd): split Execute into Build with IO/Keychain injection

Introduce a public cmd.Build entry point so external consumers (cli-server,
MCP server, other embedders) can assemble the full CLI command tree without
going through os.Args or the platform keychain. Build takes an
InvocationContext plus functional BuildOptions:

  * WithIO(in, out, errOut) — inject custom streams; terminal detection
    is derived from the input's underlying *os.File when present.
  * WithKeychain(kc)        — swap the credential store.
  * HideProfile(bool)       — registered later in cmd.HideProfile.

The existing Execute() keeps using the internal buildInternal (which
still returns the Factory so error handling can attribute exit codes),
and SetDefaultFS replaces the global VFS implementation at startup.

Hardening applied up front:

  * cmdutil.NewIOStreams(in, out, errOut) centralizes terminal detection
    so SystemIO() and WithIO share one path.
  * cmdutil.NewDefault normalizes partial IOStreams — callers may pass
    &IOStreams{Out: buf} without tripping nil-writer panics in the
    RoundTripper warnings, Cobra, or the credential provider.
  * Build guards against nil functional options.
  * An API contract test (cmd/build_api_test.go) exercises Build +
    WithIO + WithKeychain + HideProfile + SetDefaultFS so the public
    surface is reachable by deadcode analysis.

Change-Id: I7c895e6019817401accbde2db3ef800da40ad319

* feat(schema): filter methods by strict mode in schema output

When strict mode is active, schema output now excludes methods that
are incompatible with the forced identity. This applies to both
pretty and JSON output formats at the resource and method levels.

Change-Id: I39647d5578466c3e23dc545bfb917ae075203ad7

* refactor: centralize strict-mode as flag registration

Change-Id: Iec11151c5002c2f58a8aa067d08747db2e4d2d8c

* fix(cmd): align strict-mode completion and build context; drop dead register shims

Thread a context.Context through RegisterShortcuts, RegisterServiceCommands,
and service.registerService/Resource/Method by introducing explicit
*WithContext variants. Pass that context into NewCmdServiceMethodWithContext
so shortcut and service command construction can honor cancellation and
strict-mode pruning consistently.

Also drop the context-less registerMethod and registerResource shims —
they became unreachable once the WithContext variants took over, and
were the source of new deadcode warnings. registerService is retained
because service_test.go still calls it directly.

Change-Id: I3fe5673aed663c7383bbbc5b0ae94d1f3491f22d

* refactor(cmd): hide --profile in single-app mode via build option

- GlobalOptions gains HideProfile; RegisterGlobalFlags stays pure and reads
  the policy off the struct. No boolean-trap parameter, one call per site.
- buildConfig holds GlobalOptions inline so HideProfile(bool) BuildOption
  mutates it directly. buildInternal stays a pure assembly function and
  requires callers to supply WithIO — no implicit os.Std* fallback.
- Add WithIO BuildOption (wrapping raw io.Reader/Writer with automatic
  *os.File TTY detection); Execute injects streams explicitly and decides
  profile visibility via HideProfile(isSingleAppMode()).
- installTipsHelpFunc force-shows hidden root flags while rendering the
  root command's own help, so single-app users still discover --profile
  via lark-cli --help without it polluting subcommand helps.

Change-Id: I7755387e993992ca969e0a4a6f54441cc1993eef

* feat(transport): extension abort hook and shared base transport

Two transport-layer changes bundled because both reshape the base
round-tripper contract used by the HTTP client, the Lark SDK client,
and the in-process updater.

1. Extension abort hook (PreRoundTripE).

   Extensions implementing exttransport.AbortableInterceptor can now
   return an error from PreRoundTripE to skip the built-in chain. The
   post hook still fires with (nil, reason) so extensions can unwind
   resources. extensionMiddleware captures the provider name so the
   returned *AbortError carries attribution.

2. Shared base transport to stop RPC leak.

   util.NewBaseTransport cloned http.DefaultTransport on every call, so
   each cmdutil.Factory produced a fresh *http.Transport whose
   persistConn readLoop/writeLoop goroutines lingered until
   IdleConnTimeout (~90s). Invisible in a single-process CLI, but the
   fork is consumed by cli-server where each RPC request constructs a
   new Factory, causing linear memory + goroutine growth under load.

   Replace NewBaseTransport with SharedTransport — returns
   http.DefaultTransport (the stdlib-wide singleton) by default, and
   a cached proxy-disabled clone only when LARK_CLI_NO_PROXY is set.
   Return type is http.RoundTripper to discourage in-place mutation of
   the shared instance. FallbackTransport is kept as a thin
   *http.Transport wrapper so existing callers in internal/auth and
   internal/cmdutil transport decorators (which were already on the
   singleton path) do not have to migrate.

   Leak-site migrations: factory_default.go (HTTP + SDK base) and
   update.go now call SharedTransport directly.

Change-Id: Ia82462134c5c5ee838be878b887860f41446a235

* fix: unblock Build() zero-opts path and sidecar demo build

Two regressions surfaced on refactor/build-execute-split:

1. cmd.Build(ctx, inv) without WithIO panicked at rootCmd.SetIn/Out/Err
   because cfg.streams stayed nil — NewDefault normalized internally
   but cmd/build.go never saw the normalized value. Default cfg.streams
   to cmdutil.SystemIO() before the root command wires them, and add a
   TestBuild_NoOptions regression guard.

2. sidecar/server-demo/main.go still called cmdutil.NewDefault(inv),
   so `go build -tags authsidecar_demo ./sidecar/server-demo` failed
   with "not enough arguments". Pass nil for the new streams parameter
   to preserve the prior behavior (NewDefault substitutes SystemIO).

Change-Id: I20227b2355cde7d19e22eba3eb841c6d8611e8a7
2026-04-21 14:48:40 +08:00