zju3dv/supernav/global-navigation

Claude Code and agent skill by zju3dv/supernav

Trust tier 3
  • community
Source check pending

Checking that the SKILL.md file is still in the repository.

Mapless long-horizon Habitat-GS navigation for global tasks. Use when an episode requires searching across rooms, maintaining target evidence, choosing visible proxies, auditing local branches, recovering from stalls, and stopping only on final visual evidence.

Install

npx vskill@latest install zju3dv/supernav/global-navigation

SKILL.md

Global Navigation

Use MUST, MUST NOT, SHOULD, and MAY literally. Rule IDs are stable audit labels.

Load the Needed Rules

  • Before the first junction registration or reuse of a branch_id, read

references/spatial-memory.md completely.

  • Before leaving the local area, taking a long return, or acting on a frontier warning,

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

  • GN-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.

  • GN-LIFE-002 — Grounding rules. MUST read the complete run-local

locate-anything/SKILL.md before the first hab_visual_ground_preview.

  • GN-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.

  • GN-LIFE-004 — Formal finish. MUST obtain exactly one successful close result with

closed=true, then answer. An audit challenge is not a close and MUST be resolved as specified under GN-TERM-002.

Target and Observation Gate

  • GN-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 or detector result alone is insufficient; a detector miss does not prove absence.

  • GN-OBS-002 — Observe–decide–act. After init and every successful movement 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.

  • GN-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=...

Current pixels outrank bounded memory. An empty ledger frontier is not house exhaustion. A movement that empties frontier may expose the next junction. MUST NOT close blocked or start a return in the same decision as a successful movement; first report the fresh surround with GN-OBS-003.

Core Navigation Decisions

  • GN-MEM-001 — Register real choices. When one latest surround contains two to four

distinct new/uncertain sibling transitions, MUST register all of them before selecting one. With one new transition, inspect it directly. Read the spatial-memory reference before registration or branch reuse.

  • GN-MEM-002 — Branch meaning. A branch_id records one observed opening and an audited

movement status. It is not a room label, candidate ID, off-screen waypoint, geometric threshold proof, or proof of exploration. When a branch-bound grounding reports entered but current pixels still show the doorframe or threshold, treat the branch operationally as approached and retain its ID and phrase in compact working frontier until pixels prove a real crossing or meaningful inspection. MUST reuse an ID only after exact visual re-observation of that opening and sibling context, and MUST pass it on both preview and confirmation.

  • GN-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 frontier. A clearly mismatched room MAY be deferred while a stronger route exists, but MUST remain remembered as frontier. MUST NOT call an unfamiliar opening exhausted only because its guessed room type seems unlikely.

  • GN-MOVE-002 — Visible bounded movement. MUST ground only a target or proxy visible in

the selected latest view. SHOULD use LocateAnything for named objects, furniture, doorways, corridors, and semantic proxies; MAY use visual point navigation for bounded threshold crossing, open-floor adjustment, framing, or recovery. Reaching a proxy is not completion. Any movement, turn, or fresh panorama expires old image refs, points, markers, and action tokens.

  • GN-MOVE-003 — Long-move gate. Before any planned move of at least 3 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, or targets an exact unresolved frontier, provided no more relevant local choice or unregistered opening is skipped. Record lower-relevance siblings for later. Read the exploration reference before confirming.

Stop Gates

  • GN-TERM-001 — Achieved. MUST close achieved only when the latest surround shows the

complete target, distinguishing attributes, required relations, and matching context at human-verifiable distance.

  • GN-TERM-002 — Blocked. MUST read the recovery reference and exhaust its bounded gate

before blocked close. If the first blocked close returns blocked_close_audit_required, the session is still open. MUST either continue exploring (which invalidates that token) or immediately retry with the returned token, an exact audit for every reported frontier branch, and any required post-entry exception. MUST NOT answer until the 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

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
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/zju3dv/supernav/global-navigation/badge)](https://verified-skill.com/skills/zju3dv/supernav/global-navigation)

Something wrong with this skill or its page? Tell us.