Files
hyperguild/ingestion/internal/mcp/tools_capture.go
mathiasandClaude Opus 5 36b757fabf
CI / Lint / Test / Vet (pull_request) Successful in 31s
CI / Mirror to GitHub (pull_request) Skipped
feat(mcp): add brain_read + brain_edit, contained to the sync source (brain#4)
brain_read returns a note verbatim; brain_edit is str_replace with an
exactly-once match. Both operate on INGEST_BRAIN_DIR, which is the pod's
mount of /var/lib/supervisor/brain — the tree brain-sync commits from, so
an edit reaches gitea on the next cycle rather than landing somewhere the
sync cannot see it (brain#2).

An adversarial design review before implementation changed three things,
each of which would otherwise have shipped:

1. Governance files are NOT in the model-facing tool. The original plan
   allowed root *.md so CLAUDE.md had an MCP path at all — the absence of
   one is what pushed a session to edit via the gitea REST API on
   2026-06-16 and wedge the sync. But brain content is ingested from the
   outside world, reaches a model's context via brain_query, and that
   model would hold the editor: one poisoned page could rewrite the
   CLAUDE.md a .dispatch-allow agent later loads as instructions. Worse,
   brain-sync merges with --strategy-option=ours, so the live tree wins
   against a human's corrective push on exactly the conflicting lines.
   CLAUDE.md now goes through brain_edit_root: de-surfaced from
   tools/list (the brain_index precedent, ADR-0012 rule 4) and gated on
   the local static principal, so a claude.ai-origin session is refused.
   The two scopes are disjoint — the privileged door has the smaller reach.

2. Reads needed the same allowlist, and brain_get already lacked it.
   api.ReadNote resolves any brain-relative path and parseFrontmatter
   returns the whole file when there is no fence, so brain_get reads
   verbatim session transcripts under archive/ (up to 500 KB, never
   BM25-indexed, so nobody thinks of them as reachable). A test proves
   the exposure was live; the same Resolve now gates brain_get.

3. Canonicalise before allowlisting. A prefix check on the raw argument
   accepts knowledge/../.git/hooks/post-commit, which Clean collapses to
   a git hook that brain-sync executes inside a pod holding the Gitea
   push credentials. Traversal, symlink escape, dotfiles, .git/, archive/,
   sessions/ and raw/ are refused by a static backstop that does not
   depend on the allowlist being right. There were no traversal tests in
   this repo before.

Also from the review: overlapping match counting (bytes.Count says "aa"
occurs once in "aaa", so a Count-based uniqueness check accepts an
ambiguous edit); UTF-8 validation on both sides, because encoding/json
substitutes U+FFFD and would hand back bytes the file does not contain,
making every derived old_str unmatchable; refusal while .git/MERGE_HEAD
or index.lock exists, since an edit racing brain-sync's merge is either
hard-reset away or committed as a wrongly resolved conflict; flock plus a
per-path mutex with the read taken inside the lock, because two pods
share the hostPath during a RollingUpdate and uniqueness evaluated
against a stale read loses an update while reporting success; same-dir
*.tmp + rename + chmod 0644 + fsync of file and parent; and reindex only
when the edit touches frontmatter, since BM25 walks live and embeddings
are mtime-driven.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BeAp5LTscnz2W7Ubt4eJ6m
2026-08-18 17:01:37 +02:00

102 lines
4.1 KiB
Go

package mcp
import (
"context"
"encoding/json"
"fmt"
"github.com/mathiasbq/hyperguild/ingestion/internal/capturehttp"
)
// principalKey is the context key under which the authenticated principal
// (re-derived in ServeHTTP) is stashed for the capture tool.
type principalKeyT struct{}
var principalKey principalKeyT
type principalInfo struct {
principal string
viaStatic bool
}
func withPrincipal(ctx context.Context, principal string, viaStatic bool) context.Context {
return context.WithValue(ctx, principalKey, principalInfo{principal: principal, viaStatic: viaStatic})
}
// captureToolDescriptor is the tools/list entry for the capture relay.
// Appended only when WithCapture has wired the tool.
func captureToolDescriptor() map[string]any {
str := func(d string) map[string]any { return map[string]any{"type": "string", "description": d} }
insightItem := map[string]any{
"type": "object",
"properties": map[string]any{
"text": str("the insight body"), "wing": str("brain wing"),
"hall": str("brain hall (facts/decisions/failures/hypotheses/sources)"),
"supersede_slug": str("optional: slug of a prior note to revise in place instead of creating"),
},
"required": []string{"text", "wing", "hall"},
}
ticketItem := map[string]any{
"type": "object",
"properties": map[string]any{
"repo": str("gitea repo (owner is always mathias)"), "action": str("create|close|comment"),
"number": map[string]any{"type": "integer", "description": "issue number (close/comment)"},
"title": str("issue title (create)"), "body": str("issue/comment body"),
},
"required": []string{"repo", "action"},
}
schema := map[string]any{
"type": "object",
"properties": map[string]any{
"context": map[string]any{
"type": "object",
"properties": map[string]any{
"harness": str("descriptive harness label (telemetry only, never a gate input)"),
"session_ref": str("optional session reference"), "fidelity": str("live-capture|transcript-parse|agent-runlog"),
"actor": str("acting user/agent"), "classification": str("caller-declared sensitivity: public|internal|confidential"),
},
},
"insights": map[string]any{"type": "array", "items": insightItem},
"tickets": map[string]any{"type": "array", "items": ticketItem},
"dry_run": map[string]any{"type": "boolean", "description": "validate + return the would-be receipt, write nothing"},
},
}
b, _ := json.Marshal(schema)
return map[string]any{
"name": "capture",
"description": "Persist a session's value uniformly: insights → brain (write or supersede), action items → Gitea tickets. The relay door for MCP-native harnesses. Origin is server-derived from your authenticated identity; confidential captures through a us-nexus surface are refused (I1). Returns a partial-aware receipt.",
"inputSchema": json.RawMessage(b),
}
}
// brainCapture is the MCP capture tool: the #55 relay for MCP-native
// harnesses. It re-uses the same CaptureService, principal-derivation, and
// origin resolver as POST /capture — only the transport differs. It holds
// no state and retains nothing beyond the I5 audit record.
func (s *Server) brainCapture(ctx context.Context, args json.RawMessage) (json.RawMessage, error) {
if s.capture == nil {
return nil, fmt.Errorf("capture tool not configured")
}
info, ok := ctx.Value(principalKey).(principalInfo)
if !ok || info.principal == "" {
// No authenticated principal ⇒ cannot derive origin ⇒ cannot gate.
return nil, fmt.Errorf("capture requires an authenticated principal")
}
in, err := capturehttp.DecodeRequest(args)
if err != nil {
return nil, fmt.Errorf("invalid capture request: %w", err)
}
// Principal and origin are server-derived — never taken from the body.
in.Context.Principal = info.principal
in.Context.Origin = s.capture.resolver.Resolve(info.principal, info.viaStatic)
rec, err := s.capture.svc.Capture(ctx, in)
if err != nil {
// Surface I1/I5 refusals and validation failures verbatim; errors.Is
// markers (ErrSovereigntyRefused / ErrAuditUnavailable) ride in the message.
return nil, err
}
return json.Marshal(rec)
}