zju3dv/supernav/multi-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.
Multi-goal mapless long-horizon Habitat-GS navigation driven by the geometry-based visual-point local walker (hab_visual_point_navigate). Use when an episode lists several ordered or unordered find goals and each reached goal must be marked in the hab_nav_goals ledger before the final close. Extends global-navigation-geo-based-executor with the goal lifecycle; all of its navigation rules still apply.
Install
npx vskill@latest install zju3dv/supernav/multi-global-navigation-geo-based-executorSKILL.md
Multi-Goal Global Navigation (Geo-Based Executor)
Use MUST, MUST NOT, SHOULD, and MAY literally. Rule IDs are stable audit labels.
Base Skill
- MG-BASE-001 — Inherited navigation policy. MUST read the complete run-local
global-navigation-geo-based-executor/SKILL.md before scene action and follow all NM- rules (lifecycle, observation gate, movement decisions, recovery, stop evidence) for the rest of the episode. Read its `references/` files when it tells you to. This skill only adds the multi-goal lifecycle; it never relaxes an NM- rule.
- MG-BASE-002 — Walker rules. NM-LIFE-002 still applies: MUST read the complete
run-local localnav-pointnav-geo-based-executor/SKILL.md before the first hab_visual_point_navigate.
Read these from the same run-local skill bundle; never substitute a user-level copy.
Goal Ledger
- MG-LIFE-001 — Reconcile after init. Immediately after the one successful
hab_init_scene, MUST call hab_nav_goals(action="status") and confirm the listed goal indexes and descriptions match the mission list. The ledger is the only authoritative progress record. MUST NOT track goal progress in files, shell, or any side channel (NM-LIFE-003 applies to progress notes too).
- MG-LIFE-002 — Re-reconcile on doubt. After context loss, a long recovery, or any
uncertainty about which goals are found, MUST call hab_nav_goals(action="status") again before deciding the next move.
Active Goal Discipline
- MG-GOAL-001 — One active hypothesis. Work on exactly one pending goal at a time.
NM-OBS-001's complete target hypothesis applies to the active goal only. When the mission states an order (First/Then/Finally), MUST pursue goals in that order.
- MG-GOAL-002 — Switch only on a successful mark. MUST NOT change the active goal
until the current one is marked found. A failed mark leaves the active goal unchanged. Incidentally seeing a later goal MAY be noted, but MUST NOT trigger a switch while the active goal is pending.
Mark Gate
- MG-MARK-001 — Evidence bar. MUST mark only when the active goal meets the
NM-TERM-001 standard: the latest surround shows the complete target, its distinguishing attributes, required relations, and matching context at human-verifiable distance, with the agent already at the floor in front of it. Seeing the target across a room is not markable; hop close first.
- MG-MARK-002 — Reconcile before marking. MUST call
hab_nav_goals(action="status") in the same decision as the intended mark, confirm the target index is still pending and its description matches the current visual evidence, then call hab_nav_goals(action="mark", target_index=k) with that exact index. Marks cannot be revoked; a wrong mark is permanent.
- MG-MARK-003 — Mark result handling. A mark without a successful result is not a
mark. MUST resolve the returned error (unknown index, already found, or missing ledger) before continuing; MUST NOT blind-retry the same failing mark.
Stop Gates
For intermediate goals, apply the base skill's final visual approach before marking, then continue to the next goal. Close the session only after the complete mission or bounded recovery failure. The ledger and approach heuristics are agent evidence; the public benchmark computes success independently from the recorded trajectory.
- MG-TERM-001 — Achieved. MUST close achieved only when a fresh
hab_nav_goals(action="status") shows every goal found and the latest surround still shows the final goal at NM-TERM-001 standard. Then close exactly once with hab_close_session(outcome="achieved").
- MG-TERM-002 — Blocked. If any goal stays pending after the bounded recovery gate
(read global-navigation-geo-based-executor/references/recovery-and-stop.md first), MUST close blocked instead of marking speculatively or closing achieved with goals pending. A blocked close still requires the recovery reference's audit with every clause 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/multi-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/multi-global-navigation-geo-based-executor)