feat: online verification, node removals, and knowledge browser

Adds claim-bound online evidence records and citation verification scripts,
node removal transport/adapter, retrieval benchmark and owner-review tooling,
GraphCanvas/SourceDetail knowledge browser UI, and the ingest handoff helper.
Bumps the agents/gewu pack to the online-verification branch.

scripts/prepare_ingest_handoff.py

@@ -0,0 +1,68 @@
+"""Print an explicit, bounded operator handoff without reading credentials or calling GEWU."""
+
+from __future__ import annotations
+
+import argparse
+import re
+
+
+def identifier(value: str) -> str:
+    if not re.fullmatch(r"[A-Za-z0-9][A-Za-z0-9._:-]{0,127}", value):
+        raise argparse.ArgumentTypeError("use a non-empty identifier of at most 128 characters")
+    if value.startswith("gwu_"):
+        raise argparse.ArgumentTypeError("supply a non-secret identifier, not a bearer credential")
+    return value
+
+
+def tool_budget(value: str) -> int:
+    try:
+        budget = int(value)
+    except ValueError as error:
+        raise argparse.ArgumentTypeError("tool budget must be an integer") from error
+    if not 1 <= budget <= 200:
+        raise argparse.ArgumentTypeError("tool budget must be between 1 and 200")
+    return budget
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("--principal-id", required=True, type=identifier)
+    parser.add_argument("--project-id", required=True, type=identifier)
+    parser.add_argument("--file-id", required=True, action="append", type=identifier)
+    parser.add_argument("--resume-run-id", type=identifier)
+    parser.add_argument("--max-tool-calls", default=35, type=tool_budget)
+    args = parser.parse_args()
+    run_instruction = (
+        f"Resume run {args.resume_run_id}: call run_status first, "
+        "preserve its run/source identities "
+        "and unchanged pack pins, and do not create a replacement run."
+        if args.resume_run_id
+        else "After diagnostics and project context, create the service run using the installed "
+        "compact run_pins before transferring source bytes."
+    )
+    print(
+        "Use the installed GEWU kb-ingest skill. Read its profile, contribution/evidence/scope "
+        "protocols and ingestion pipeline once. Use discovered MCP schemas and the compact "
+        "factual draft contract in CONTRIBUTIONS.md.\n\n"
+        f"Operator-supplied principal_id={args.principal_id}; project_id={args.project_id}; "
+        f"selected connector file_ids={', '.join(args.file_id)}; "
+        f"turn budget={args.max_tool_calls} tool calls. "
+        "This identifier supplies attribution, not additional authority; the service credential "
+        "must authenticate it. Use it for proposer_id, agent_id and assessor_id.\n\n"
+        f"{run_instruction}\n\n"
+        "Process one small evidence-backed learning unit across the selected sources this turn. "
+        "Recover their connector receipts, read relevant bounded extraction slices, search for "
+        "duplicates, prepare a complete draft, evaluate, and submit only eligible work. Pass "
+        "prepared proposal_json unchanged. Keep policy holds and unsupported claims explicit. "
+        "Checkpoint actual coverage, accepted receipts and pending sections before returning. "
+        "Do not claim the paper or packet complete from this bounded turn.\n\n"
+        "Do not inspect credentials, environment variables, host MCP configuration, service "
+        "databases or old benchmark logs. Never probe proposal tools with empty objects or "
+        "guessed identities. Correct one actionable argument error; stop an unchanged failure "
+        "loop and report the exact blocker. Respect the tool budget and return a partial "
+        "checkpoint/report if work remains."
+    )
+
+
+if __name__ == "__main__":
+    main()