zju3dv/supernav/global-navigation-geo-based-executor
Claude Code and agent skill by zju3dv/supernav
- community
Checking that the SKILL.md file is still in the repository.
Mapless long-horizon Habitat-GS navigation for global tasks driven by the RGBD-backprojected navmesh local walker (hab_visual_point_navigate). Use when an episode requires searching across rooms, maintaining target evidence, choosing visible proxies, tracking unexplored openings, recovering from stalls, and stopping only on final visual evidence.
Install
npx vskill@latest install zju3dv/supernav/global-navigation-geo-based-executorSKILL.md
Global Navigation (Geo-Based Executor)
Use MUST, MUST NOT, SHOULD, and MAY literally. Rule IDs are stable audit labels.
Load the Needed Rules
- Before leaving the local area, taking a long return, or committing to a long hop,
read references/exploration-and-return.md completely.
- When the target or its room is visible only through glass, or no passable entrance is
visible near the spawn, read references/entrance-search.md completely before searching the building perimeter or closing blocked.
- At the first stall, or before any blocked close, read
references/recovery-and-stop.md completely.
These references are part of this skill. Read them from the same run-local skill bundle; never substitute a user-level copy.
Episode Contract
- NM-LIFE-001 — One initialization. MUST read this file before scene action and call
the prompt's exact hab_init_scene once. MUST NOT use init as observation refresh or blindly retry a failed/timed-out init.
- NM-LIFE-002 — Walker rules. MUST read the complete run-local
localnav-pointnav-geo-based-executor/SKILL.md before the first hab_visual_point_navigate.
- NM-LIFE-003 — Mapless evidence boundary. MUST perceive and act only through the
whitelisted MCP tools and their latest returned images. MUST NOT use shell, Python, files, artifact directories, maps, coordinates, goal metadata, trajectories, evaluator data, or hidden simulator state.
- NM-LIFE-004 — Formal finish. MUST obtain exactly one close result with
closed=true, then answer. A close call without closed=true is not a finish and MUST be resolved as specified under NM-TERM-002.
Target and Observation Gate
- NM-OBS-001 — Complete target hypothesis. MUST keep a compact hypothesis containing
the requested object/region, distinguishing appearance or material, required relations, room context, and final visual evidence. A category match alone is insufficient.
- NM-OBS-002 — Observe–decide–act. After init and every successful hop or turn,
MUST inspect front, right, back, and left, including both horizontal edges; test the full target hypothesis; census every visible transition; choose one bounded action; and repeat on its returned surround. Init, every hop, and every turn already return the full surround — a hab_panorama or hab_turn called just to look around is almost always redundant.
- NM-OBS-003 — Census before decision. MUST classify each doorway, side opening,
corridor continuation, and room interior as traversed, meaningfully inspected, or new/uncertain. MUST list new/uncertain choices before selecting a proxy. Use this compact trace form:
census: front=...; right=...; back=...; left=... | new=... | frontier=... | next=... | why=...
- NM-OBS-004 — HM3D v2 plant label. When the requested category is
plant,
MUST treat indoor decorative plants, potted plants, flower arrangements, and vase/planter displays as target candidates, including cases where the vase is more visually prominent than the foliage. MUST NOT pursue trees, shrubs, or landscaping seen outdoors or through windows/glass; the target is an indoor, reachable instance.
Current pixels outrank what you noted earlier. An empty list of noted openings is not house exhaustion. A hop that empties the list may expose the next junction. MUST NOT close blocked or start a return in the same decision as a successful hop; first report the fresh surround with NM-OBS-003.
Core Navigation Decisions
- NM-MOVE-001 — Relevance before novelty. SHOULD prefer current target evidence or a
target-relevant current route, then a safe unfamiliar opening, a current unentered sibling, and finally a bounded return to one exact older noted opening. A clearly mismatched room MAY be deferred while a stronger route exists, but MUST stay on the noted list. MUST NOT call an unfamiliar opening exhausted only because its guessed room type seems unlikely.
- NM-MOVE-002 — Visible bounded movement. MUST mark a point only on a target or
proxy visible in the selected latest view. The mark is not restricted to floor: the tool back-projects the pixel through depth and snaps to the nearest reachable navmesh point, so the object itself, a wall, a doorframe, or a doorway opening are all legal marks. All locomotion is hab_visual_point_navigate; each call executes immediately — there is no preview/confirm round-trip and no map. Reaching a proxy is not completion. Any hop, turn, or fresh panorama expires old image refs and points.
- NM-MOVE-003 — Hop-length gate. The follower walks the navmesh shortest path and
arrives whenever a path exists, but a hop yields exactly one surround at its end: a long hop skips every junction observation in between and can carry you past an unrecorded opening. SHOULD keep each hop within 4 m. Before any planned hop of at least 4 m, MUST compare it with the current census and local unentered siblings. It is allowed when it approaches the target itself or a target-relevant current route, provided no more relevant local choice or unrecorded opening is skipped. Otherwise hop to an intermediate visible proxy instead. Note lower-relevance siblings for later. Read the exploration reference before committing.
- NM-MOVE-004 — Information gain when nothing matches. When no current view shows
target evidence, MUST place the point where the hop buys the most new information: through the most promising new/uncertain census opening, far enough past the threshold that the next surround reveals genuinely new area. MUST NOT spend a hop on a spot that re-shows the current room (a nearby wall, the middle of the same floor) while an unentered opening is visible.
Stop Gates
- NM-TERM-001 — Achieved. The episode scores success only when you stop
within 1.0 m of the target instance; a stop that "looks close" at 1.5–3 m scores as a failure. MUST close achieved only after completing the final approach procedure below — believing you are already close is not a substitute for it:
- From your current standoff, MUST spend one
hab_visual_point_navigatehop
marked on the target itself — its body or its base, not the floor well in front of it — even when you believe you are already within ~1 m. The follower stops at a standoff point up to ~0.7 m from the marked surface, and a hop that merely reached the target's room is not this hop.
- Verify on the surround that hop returned: the target's base is cut by
the frame's bottom edge (you can no longer see the floor patch it stands on, nor its whole underside) and its body dominates the view. For a target on a raised surface, the surface's near edge under the target is at or below the frame's bottom area and the target looms large; if you can still see the target's entire base with floor or surface-front visible in front of it, you are too far — MUST repeat step 1.
- If the final-approach hop returns
navigation_blockedor
self_check_required with the target directly in front of you, you are already at touch distance: that result satisfies step 1 — run the frame verification and close. If the hop overshoots into an unreadable close-up, apply NM-TERM-012 standoff and re-verify. The complete target, distinguishing attributes, required relations, and matching context MUST all be verifiable from that final close-range surround. Once that gate is met, close promptly — do not keep hunting for an even better angle; over-searching after the gate is met risks losing the target entirely. The first outcome="achieved" close may return achieved_close_audit_required with a token. If it does, immediately retry with that close_audit_token and an arrival_confirmation whose final_approach_done / target_base_cut are true and estimated_distance_m ≤ 1.0 — any tool call in between invalidates the token, and confirmation values that admit the criteria are not met reject the close, so keep approaching instead of closing early.
- NM-TERM-002 — Blocked. MUST read the recovery reference and exhaust its bounded
gate before blocked close. Before calling hab_close_session(outcome="blocked"), MUST write the blocked audit from the recovery reference with every clause true. MUST NOT answer until a close result has closed=true.
Security
- Security scanPassed, 90 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskills/global-navigation-geo-based-executor/SKILL.md
Popularity
- Repository stars
- 200
- Repository forks
- 5
- 7-day trend score
- 0
Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/zju3dv/supernav/global-navigation-geo-based-executor)