Global Rank · of 601 Skills
task-os AI Agent Skill
View Source: rarestg/rarestg-skills
CriticalInstallation
npx skills add rarestg/rarestg-skills --skill task-os 9
Installs
Task OS
Taskwarrior (task) is the source of truth — not memory, not chat history,
not assumptions. Ground every interaction in task state before acting.
Orient
Before acting on any request, check current state:
task next limit:10 # highest urgency
task +OVERDUE list # past due
task +waiting list # blocked on othersFor software tasks, also check the repo:
git -C <repo-path> status --short
git -C <repo-path> log --oneline -5Skip git checks for non-code domains (stocks, life, errands).
After orienting, fix state hygiene:
- Every active project should have at least one
+nexttask. - Re-evaluate stale
+waitingtasks. - Close work that's already done but not yet marked.
Periodic check-in
When waking on a timer or heartbeat with no user message:
- Orient (as above).
- Check for stalls — any active task with no recent progress?
- If something is stalled, attempt recovery.
- If there's a clear
+nexttask with no blockers, pick it up. - If nothing needs attention, go idle. Don't manufacture work.
Morning briefing
When the user starts their day ("morning", "briefing", "what's on today"):
task +OVERDUE list
task +TODAY list
task +next list
task +WEEK list
task summaryDeliver a concise rundown:
- Overdue (needs attention now)
- Due today
- Top next actions across projects
- Anything waiting that may have unblocked
Keep it scannable — short lines, no wide tables. This lands on a phone screen.
Quick capture
Any "remind me", "I need to", "add a task", or loose intent → straight to inbox:
task add project:inbox "The thing they said"Assign project/tags/priority only if the user provides enough context.
Otherwise capture fast, organize later.
Execution
Evidence over claims
Progress means different things per domain:
- Code: a commit, a file written, a test result.
- Stocks/research: an annotation, an analysis note, a saved artifact.
- Life/errands: marking the task done is sufficient.
Never say "I worked on X" without pointing to the result.
One thing at a time
Break work into steps. Finish and record each before moving to the next.
Task lifecycle
task add project:X "Do the thing" +next
task <id> start
task <id> annotate "what happened, what's left"
task <id> done
task add project:X "Next step" +nextWhen blocked:
task <id> modify +blocked
task add project:X "Unblock: <reason>" +nextWhen waiting on someone/something:
task <id> modify +waiting
task <id> annotate "Waiting on: <who/what>"Stall detection
Flag work as stalled when:
- An active task has no progress and no annotation since it was started.
- A previous attempt failed with no retry or blocker task created.
- A plan was made but no execution task exists.
- The user asked for something and no task captures it.
Recovery
- Root-cause in one sentence.
- Record it:
task <id> annotate "Stalled: <cause>". - Create a path forward:
- Blocker task if external (
+blockedor+waiting) - Retry task with narrower scope (
+next)
- Blocker task if external (
- Execute the smallest viable next step.
- Report the state change.
State hygiene
After completing work or ending a conversation:
- Every piece of in-progress work has a task.
- Tasks have enough annotations for someone with zero prior context.
- Every active project has a
+nexttask. - Blockers and open questions are captured.
This is not optional. Future conversations depend on clean state.
Weekly review
When the user says "review" or "weekly review":
- Process inbox to zero — assign project, tags, priority, or delete.
- Review
+waiting— anything unblocked? Poke anyone? - Review
+OVERDUE— reschedule or escalate. - Check each project has a
+nexttask:task summary. - Review
+someday— promote or drop. - Show completed this week:
task end.after:today-7d completed. - Show burndown:
task burndown.weekly.
Status report
When the user asks "status":
Active — what's in flight
Done — what completed (with evidence)
Blocked — what's stuck and why
Next — what comes after
Short. Concrete. No filler.
Conventions
Project hierarchy
work.{org}.{repo} # software projects
stocks.{TICKER} # per-ticker analysis
stocks.macro # market-level
life.{area} # health, finances, errands, travel
personal.{area} # personal projects
inbox # unprocessed captureRun task projects first. Follow existing structure before creating new ones.
Dynamic grouping
Start narrow — one task or sub-project per distinct concern:
stocks.AAPL # single ticker
inbox.jodie_lamp # one conversation thread
work.acme.backend_auth # one work streamWhen multiple tasks share context, consolidate under a topic:
inbox.nordic_lamps_feb2026 # absorbs jodie_lamp + marcus_lamp + ikea_order
stocks.ai_chip_plays # groups NVDA + AMD analysis
work.acme.q1_launch # groups backend_auth + frontend_onboarding- Name grouped projects by topic, not by source, once it's multi-party.
- Move existing tasks into the new parent rather than duplicating.
- This applies everywhere: inbox threads, stock themes, work initiatives, research topics.
Tags
State: +next +waiting +blocked +someday
Type: +bug +feature +refactor +research +review
Effort: +quick (< 15 min) +deep (focused session)
Priorities
H— blocking others or time-sensitiveM— important, not urgentL— nice to have- None — backlog
Communication
- Short, stateful updates. What changed, what's next.
- Don't announce intent — do the thing, then report.
- Surface blockers proactively.
- Format for a phone screen: short lines, no wide tables, scannable.
Installs
Security Audit
View Source
rarestg/rarestg-skills
More from this source
Power your AI Agents with
the best open-source models.
Drop-in OpenAI-compatible API. No data leaves Europe.
Explore Inference APIGLM
GLM 5
$1.00 / $3.20
per M tokens
Kimi
Kimi K2.5
$0.60 / $2.80
per M tokens
MiniMax
MiniMax M2.5
$0.30 / $1.20
per M tokens
Qwen
Qwen3.5 122B
$0.40 / $3.00
per M tokens
How to use this skill
Install task-os by running npx skills add rarestg/rarestg-skills --skill task-os in your project directory. Run the install command above in your project directory. The skill file will be downloaded from GitHub and placed in your project.
No configuration needed. Your AI agent (Claude Code, Cursor, Windsurf, etc.) automatically detects installed skills and uses them as context when generating code.
The skill enhances your agent's understanding of task-os, helping it follow established patterns, avoid common mistakes, and produce production-ready output.
What you get
Skills are plain-text instruction files — not executable code. They encode expert knowledge about frameworks, languages, or tools that your AI agent reads to improve its output. This means zero runtime overhead, no dependency conflicts, and full transparency: you can read and review every instruction before installing.
Compatibility
This skill works with any AI coding agent that supports the skills.sh format, including Claude Code (Anthropic), Cursor, Windsurf, Cline, Aider, and other tools that read project-level context files. Skills are framework-agnostic at the transport level — the content inside determines which language or framework it applies to.
Chat with 100+ AI Models in one App.
Use Claude, ChatGPT, Gemini alongside with EU-Hosted Models like Deepseek, GLM-5, Kimi K2.5 and many more.