210 — NagaAgent exploration: host-mediated forum publishing and useful fingerprints Recorded 2026-09-06 UTC. Published source reviewed; no code or prompt executed. RESULT NagaAgent implements internet exploration through OpenClaw and a separate host-program step that automatically publishes a condensed result to its community forum. This supplies a concrete mechanism and distinctive post-format markers. No matching live public post or execution transcript was recovered, and no escaped swarm is established. PINNED PRIMARY SOURCE https://github.com/RTGS2017/NagaAgent Tree c2caa9079b9eb48129f550c43a5485231d404d3b: 5,503 paths, non-truncated. Five downloaded source/doc files match their Git blob hashes. This validates the captures against current source, not deployment history. EXPLORATION AND LOCAL RECORDS The system document describes user-created tasks with a direction, selected operator, time/credit limits and browser settings. Chat and exploration reuse the same operator's OpenClaw instance with separate session keys; one operator is not automatically two agents because it has two sessions. Exploration session keys use travel:{agent_id}:{first-eight-session-id-characters}. Session JSON and final Markdown reports live under the user's .naga/travel directory. These are documented local paths, not public download endpoints. We did not seek anyone's private session files. Distinctive tools are travel_progress, travel_discovery, travel_state and travel_summary. The vendored tool implementation limits them to travel sessions; travel_summary writes a randomly named Markdown file locally. The prompt also supports [DISCOVERY] blocks when tools are unavailable. AUTOMATIC FORUM PUBLISHING TravelSession.post_to_forum defaults to true in apiserver/travel_service.py. In agentserver/agent_server.py, after exploration, the host program checks post_to_forum and a nonempty summary, builds a forum payload and calls create_forum_post_internal. This is a deterministic application publication step surrounding the model run. It does not require the model to independently decide to call a forum-post tool at that point. A user's configured setting can change the behavior; source defaults are not proof of any specific user's settings. The payload builder emits: - title prefix: 探索速报｜ - section markers: 【探索方向】, 【精华总结】, 【本轮值得继续追踪】 - optional social-count section: 【社交互动】 - source field: openclaw-travel - at most five discovery entries and five tags; content truncated to 4,000 characters, title to 60. This gives a better search fingerprint than generic travel or agent terminology. A matching format would indicate software compatibility, not prove operator or model identity. There is also a separate build_social_prompt function encouraging forum interaction after useful research progress and recording [SOCIAL] blocks. Its existence is not evidence that a particular run used it or successfully interacted. Do not conflate the optional model-directed social instruction with the host's automatic digest publication. POSTED STATUS IS NOT AN INDEPENDENT RECEIPT The host extracts a post ID from several possible response shapes, potentially leaving it null. It then sets forum_post_status to posted if the call did not raise. Consequently, a published local posted status without a concrete post ID and readable destination should not be treated as a complete cross-site join. Error paths record a failed status. No actual run response was inspected. ACCESS BOUNDARY The inspected forum proxy requires a Naga access token (with refresh handling) before contacting the configured NagaBusiness forum API. Without one it raises 401 locally. This documents the client path's authentication requirement; it does not prove every public page or upstream read requires authentication. No tokens acquired or used, and no guessed private endpoints or user histories accessed. A mainland IP alone would not satisfy a credential requirement. BOUNDED WEB SEARCH Queries for exact openclaw-travel outside GitHub, 探索速报 with 娜迦, and 本轮值得继续追踪 recovered no matching public Naga exploration post in returned results. Most travel results concerned ordinary tourism assistants. Search absence is not proof no posts exist or were made. NEXT EVIDENCE TO SEEK A voluntarily public trace containing a travel session key, travel tool results and forum_post_id, paired with a readable post matching that ID, would connect execution and publication. Independent archive dates would strengthen timing. Multiple operator IDs or posts still would not by themselves establish an autonomous collaborating swarm. Preserve this exact fingerprint set for later public-corpus searches. SOURCES https://raw.githubusercontent.com/RTGS2017/NagaAgent/c2caa9079b9eb48129f550c43a5485231d404d3b/docs/travel-exploration-system.md https://raw.githubusercontent.com/RTGS2017/NagaAgent/c2caa9079b9eb48129f550c43a5485231d404d3b/vendor/openclaw/src/agents/tools/travel-tools.ts https://raw.githubusercontent.com/RTGS2017/NagaAgent/c2caa9079b9eb48129f550c43a5485231d404d3b/apiserver/travel_service.py https://raw.githubusercontent.com/RTGS2017/NagaAgent/c2caa9079b9eb48129f550c43a5485231d404d3b/apiserver/routes/forum.py https://raw.githubusercontent.com/RTGS2017/NagaAgent/c2caa9079b9eb48129f550c43a5485231d404d3b/agentserver/agent_server.py LOCAL CAPTURES investigation/china/210-private/: recursive tree, system document, travel tools, service and server source, forum proxy source, per-file Git verification metadata and SHA-256 hashes. Classification: verified implementation of exploration and automatic digest publishing; no observed public swarm activity or Chinese-lab attribution.