task-compass
Skill1 starsChanged 4 months agoMITView on GitHub
Classify agent tasks into a planner profile, execution tools, safety policy, model tier, context policy, and next action. Use when deciding how a task should be planned or delegated, selecting CLI, file, mobile GUI, mobile CLI, MCP, or deployment executors, determining whether to continue, replan, refuse, or await human approval, or evaluating routing behavior across benchmark tasks.
1 other project uses a version of this.
Adopted by
Most recently taken up by
- CodyQin/ai-task-compasstheir own version, first seen Nov 15, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- CodyQin/ai-task-compass★ 2same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repositoryOne other file it gives its agents.
- AGENTS.mdAGENTS.md
The file
We split this skill into the rules it gives an agent
10 rules. About 0 are common, 6 are distinctive, and 4 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈6
Distinctive · estimate
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 24
Do not use this skill as a command generator, a tool-argument generator, or a replacement for runtime permissions.
- Distinctive: line 28
Run the bundled router from this skill directory:
- Very specific: line 40
Use the returned planned_tools, primary_executor, policy_mode, permission_behavior, next_action, and safety_guard to construct the next …
- Distinctive: line 51
Do not pass expected labels or benchmark annotations to the router.
- Very specific: line 65
Read references/routing-contract.md when integrating the JSON output, changing enum values, or investigating a route.
- Distinctive: line 67
Read references/model-card.md when reviewing training provenance, benchmark claims, licensing, limitations, or promotion gates.
- Very specific: line 71
Never downgrade confirm or refuse to act without external verifier evidence.
- Very specific: line 72
Never turn safety_guard=true off downstream.
- Distinctive: line 74
Keep explicit read-only constraints and workspace boundaries intact.
- Distinctive: line 75
Keep model predictions, deterministic signals, and the final reason visible in audit logs.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Task Compass Skill
Produce a deterministic, auditable routing decision before planning or execution. Treat the result as advisory policy input; never use it to bypass runtime permission checks.
OpenClaw Runtime Integration
Install this repository as an OpenClaw workspace skill, or copy this directory to skills/task-compass. OpenClaw loads the instructions automatically; the bundled router is invoked explicitly when a deterministic route is needed. The deprecated route-openclaw-task name remains a compatibility alias.
This project's optional LocalAuditPlanner adapter can enable automatic runtime routing with:
OPENCLAW_PLANNER_SKILL=task-compassWhen a task includes an Available tool interfaces: catalog, resolve the required tool dependency closure from the task intent and ignore unrelated distraction tools. The runtime adapter converts execution_tools into a bounded target path and passes it to OpenClaw's A*/Reflexion planner as an auditable desired-tool constraint. Candidate generation, permission decisions, risk checks, and verification remain owned by OpenClaw.
Do not use this skill as a command generator, a tool-argument generator, or a replacement for runtime permissions.
Route One Task
Run the bundled router from this skill directory:
python3 scripts/route_task.py --goal "<task>" --prettyPass the runtime permission mode when known:
python3 scripts/route_task.py --goal "<task>" --permission-mode DEFAULT --prettyUse the returned planned_tools, primary_executor, policy_mode, permission_behavior, next_action, and safety_guard to construct the next planner step. Preserve deny, await_human, replan, and refuse outcomes exactly.
Route Many Tasks
Provide JSONL rows containing goal and optional id and permission_mode fields:
python3 scripts/route_task.py --input-jsonl tasks.jsonl --output-jsonl routes.jsonl
python3 scripts/validate_route.py routes.jsonlDo not pass expected labels or benchmark annotations to the router.
Evaluate
Measure routing behavior without copying raw prompts into the report:
python3 scripts/evaluate_router.py --suite benchmark.json --output metrics.jsonCompare task_route_success_rate, safety_guard_accuracy, required_execution_tool_coverage, and holdout results before promoting a change. Keep the existing runtime permission engine as the final authority.
Interpret Decisions
Read references/routing-contract.md when integrating the JSON output, changing enum values, or investigating a route. Do not load assets/profile-policy-model.json into the prompt; let the script read it.
Read references/model-card.md when reviewing training provenance, benchmark claims, licensing, limitations, or promotion gates.
Apply these invariants:
- Never downgrade
confirmorrefusetoactwithout external verifier evidence. - Never turn
safety_guard=trueoff downstream. - Route low-confidence actionable tasks to
replanwith expanded context. - Keep explicit read-only constraints and workspace boundaries intact.
- Keep model predictions, deterministic signals, and the final reason visible in audit logs.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

