Files
archie/internal/infrastructure/captureintake/receiver.go
T
sam a00c578cb7 refactor(ui): sever the archie-ui binary from every daemon runtime package
Completes the objective deletion gate for the UI Service
(archie-core-8cda.5.6): go list -deps ./cmd/archie-ui now links zero
banned packages -- no modernc.org/sqlite, no internal/store, gateway,
channels, memory, secret, agentexec, tools, skill, yaegiutil, curator,
or the workflow engine. The closure shrinks from 68 to 26 archie-core
packages. Each cut is a producer-side move; no consumer-side defensive
shims.

- Store contracts (TaskStore family, value types, sentinels) moved from
  internal/store to internal/domain/storecontract -- the approved contract
  location per state-store-contract rev. 2d, and NOT internal/contracts:
  that tree is buf generate output and tools/proto.sh check deletes
  hand-written files there. internal/store keeps aliases; sentinel
  message strings are byte-identical and remain the wire contract.
- Chat contracts (ChatContract, sessions, turns, streaming, commands)
  moved from internal/gateway to internal/domain/messaging; gateway keeps
  aliases. ToolCallEvent rendering helpers become gateway functions
  (methods cannot live on alias types).
- Task vocabulary (Task, statuses, mid-run Store, Definition) split into
  internal/domain/workflow/task; the pipeline engine stays in package
  workflow with aliases, so daemon-side callers are unchanged.
- overlay.DeniedKeys moved to internal/infrastructure/configuration:
  webui referenced the constant and that dragged the overlay SQLite
  store in.
- webui.Server's concrete runtime handles become webui-owned view
  interfaces with daemon-side bootstrap adapters: channel lifecycle
  (internal/channels/status), MemoryStatus, CuratorStatus, and a
  SkillCatalog seam replacing the direct skill.DefaultRoots/CatalogRoots
  calls. Each degrades to an empty page when unwired.
- config.SecretRef moved into internal/config (config vocabulary);
  internal/secret keeps an alias and Registry.Resolve absorbs the
  zero-value short-circuit.

Docs: state-store-contract rev. 2d relocation record; ui-service-boundary
deletion gate marked satisfied; ARCHITECTURE.md package map rows.
2026-09-10 03:03:33 +10:00

235 lines
9.3 KiB
Go

// Package captureintake is the HTTP receiver for inbound webhooks archie has
// no binding for yet: it persists what arrives so an operator can inspect it
// and build one (docs/prds/event-capture-storage.md).
//
// It belongs to the process that owns work intake, not to the dashboard that
// displays captures. It served from the dashboard's listener while the two
// shared a process; the split gives the read to the dashboard, over the State
// Store, and keeps the write here (archie-core-8cda.5.4). From the cutover
// change in the same bead it is a persistence shim over the State Store
// contract mounted by the UI process -- the only dashboard listener left --
// while the daemon's binding-dispatch loop consumes captures from the same
// store: the HTTP front door moved without moving work intake.
package captureintake
import (
"context"
"encoding/json"
"fmt"
"io"
"log/slog"
"net"
"net/http"
"time"
"github.com/samcharles93/archie-core/internal/domain/binding"
"github.com/samcharles93/archie-core/internal/domain/storecontract"
"github.com/samcharles93/archie-core/internal/events"
"github.com/samcharles93/archie-core/internal/webhookguard"
)
// Path is the route this receiver serves. It is unauthenticated by design:
// the senders it exists to capture have no credential here yet.
const Path = "POST /webhooks/capture/{source}"
// Receiver persists inbound webhook captures.
type Receiver struct {
// Captures persists what arrives. Nil answers 503 rather than failing
// the process it is mounted in: capture is a precondition for the
// no-code playbook epic, not a hard dependency of the daemon.
Captures storecontract.CaptureStore
// Limiter is the per-remote-address token bucket applied before a body
// is read. Nil disables rate limiting, which composition never does in
// production.
Limiter *webhookguard.RateLimiter
// Bindings resolves the armed binding whose secret authenticates a
// source. Nil disables per-source HMAC verification and every event
// records as authenticated=false.
Bindings storecontract.BindingDispatcher
// Retention and MaxEvents are passed straight to InsertCapture's
// prune-on-write bounds. See config.CaptureConfig.
Retention time.Duration
MaxEvents int
// MaxBodyBytes rejects (413) a body larger than this before redaction
// or storage sees it.
MaxBodyBytes int64
// Publish puts the capture's arrival on the activity stream, with the
// request's context: the announce happens synchronously inside the
// handler, before the 202 is written, so the request's lifetime bounds
// it. Nil records the capture without announcing it.
Publish func(ctx context.Context, e events.Event)
Log *slog.Logger
}
// Register mounts the receiver on mux.
func (rc *Receiver) Register(mux *http.ServeMux) {
mux.HandleFunc(Path, rc.ServeHTTP)
}
// ServeHTTP accepts an inbound webhook POST and persists it. If a binding is
// armed for this source the body must also carry a valid HMAC signature; only
// correctly-signed events are recorded as authenticated (t2db.5 point 1). An
// unauthenticated event is still captured, visible in the inspector, and
// marked unauthenticated -- the dispatch loop's auth check
// (binding.Matcher.Matches) is the actual gate against false task creation.
func (rc *Receiver) ServeHTTP(w http.ResponseWriter, r *http.Request) {
if rc.Captures == nil {
http.Error(w, "capture not configured", http.StatusServiceUnavailable)
return
}
source := r.PathValue("source")
// webhookguard.RateLimiter's bucket map assumes a bounded, operator-
// registered key space (see its doc comment) -- it is never evicted, and
// a fresh key starts with a full burst allowance. source is an
// unregistered, attacker-chosen URL segment by design (this endpoint
// exists to capture senders archie has no registration for yet), so
// keying on it would let a single attacker rotate a new "source" on
// every request to bypass the limit entirely and grow the bucket map
// without bound. Remote address is the bounded identity available here.
if rc.Limiter != nil && !rc.Limiter.Allow(remoteAddrHost(r)) {
http.Error(w, "rate limit exceeded", http.StatusTooManyRequests)
return
}
armed, ok := rc.armedBindings(w, r, source)
if !ok {
return
}
r.Body = http.MaxBytesReader(w, r.Body, maxBodyBytesOrFallback(rc.MaxBodyBytes))
body, err := io.ReadAll(r.Body)
if err != nil {
// The only realistic cause here is MaxBytesReader tripping; any other
// read failure also means there is no usable body to capture.
http.Error(w, "request body too large", http.StatusRequestEntityTooLarge)
return
}
// HMAC verification -- only when exactly one binding is armed for this
// source. Mirrors channels/webhook/webhook.go's header precedence:
// GitHub-style X-Hub-Signature-256 first, X-Signature-256 fallback.
// Empty header + non-empty secret is never valid (VerifyHMAC's own
// guard), so an absent signature with an armed binding lands here as
// authenticated=false rather than authenticated=true.
authenticated := false
if len(armed) == 1 {
sig := r.Header.Get("X-Hub-Signature-256")
if sig == "" {
sig = r.Header.Get("X-Signature-256")
}
authenticated = webhookguard.VerifyHMAC(body, sig, armed[0].Secret)
}
headers, _ := json.Marshal(r.Header)
redactedHeaders, err := webhookguard.RedactPayload(headers)
if err != nil {
redactedHeaders = headers
}
// A body that isn't JSON has no key-value structure for the heuristic to
// match against, so there is nothing to redact -- store it as received
// rather than dropping or mangling it. Best-effort, per
// docs/prds/webhook-intake-security.md point 5.
redactedBody, err := webhookguard.RedactPayload(body)
if err != nil {
redactedBody = body
}
c := storecontract.CapturedEvent{
ReceivedAt: time.Now().UTC(),
Source: source,
RemoteAddr: r.RemoteAddr,
ContentType: r.Header.Get("Content-Type"),
Headers: string(redactedHeaders),
Body: string(redactedBody),
Authenticated: authenticated,
}
id, err := rc.Captures.InsertCapture(r.Context(), c, rc.Retention, rc.MaxEvents)
if err != nil {
rc.logger().Error("capture insert", "err", err, "source", source)
http.Error(w, "capture failed", http.StatusInternalServerError)
return
}
// Reuses the existing activity-event pipeline rather than inventing
// separate push plumbing -- see docs/prds/event-capture-storage.md.
// Deliberately LIGHTWEIGHT: the events table this feeds has no retention
// or row-count prune, unlike captured_events, so embedding the full (up
// to MaxBodyBytes) body/headers here would duplicate the payload into an
// unbounded table and defeat InsertCapture's own disk-bound guarantee.
// The captures view treats this purely as an invalidation signal and
// refetches the list for the actual row data.
if rc.Publish != nil {
rc.Publish(r.Context(), events.Event{
Kind: "capture",
Detail: "capture from " + source,
Data: map[string]any{"id": id, "source": source},
})
}
w.WriteHeader(http.StatusAccepted)
}
// armedBindings resolves the bindings armed for source before the body is
// read. A lookup error is logged and treated as "no armed binding" -- the
// dispatch loop's auth check is the actual gate, so failing closed here would
// amplify a transient store hiccup into a capture outage for every
// sender-not-yet-bound flow. If a future phase wants fail-closed semantics,
// that is a separate decision and should not silently land here.
//
// Two armed bindings for one source would race over every inbound webhook for
// that source: the write-time overlap check should make this state
// impossible, but the dispatch loop needs a single binding per capture, so
// the overlap surfaces as 409 at capture time (belt-and-braces TOCTOU guard
// -- see the store's ApproveBinding).
func (rc *Receiver) armedBindings(w http.ResponseWriter, r *http.Request, source string) ([]binding.Binding, bool) {
if rc.Bindings == nil {
return nil, true
}
armed, err := rc.Bindings.ArmedBindingsForSource(r.Context(), source)
if err != nil {
rc.logger().Warn("armed bindings lookup", "source", source, "err", err)
return nil, true
}
if len(armed) > 1 {
http.Error(w,
fmt.Sprintf("multiple armed bindings for source %q -- overlap rejected", source),
http.StatusConflict)
return nil, false
}
return armed, true
}
func (rc *Receiver) logger() *slog.Logger {
if rc.Log == nil {
return slog.New(slog.DiscardHandler)
}
return rc.Log
}
// fallbackMaxBodyBytes bounds a capture body when MaxBodyBytes is left unset
// (zero). The size cap is the only one of the three disk-bound mechanics in
// docs/prds/event-capture-storage.md that limits a SINGLE payload rather than
// total row count or write rate -- a table pruned to maxEvents rows still
// holds one arbitrarily large row if nothing ever bounded that row's size, so
// this must never be skippable by a missing wiring value the way the rate
// limiter (best-effort, not disk-bound-load-bearing on its own) is allowed to
// be.
const fallbackMaxBodyBytes = 256 * 1024
func maxBodyBytesOrFallback(configured int64) int64 {
if configured > 0 {
return configured
}
return fallbackMaxBodyBytes
}
// remoteAddrHost strips the port from r.RemoteAddr for use as a rate-limit
// key, falling back to the raw value if it isn't a host:port pair.
func remoteAddrHost(r *http.Request) string {
host, _, err := net.SplitHostPort(r.RemoteAddr)
if err != nil {
return r.RemoteAddr
}
return host
}