~/project/.dsh/skills/ProjectLoads for this repository only. The right place to read a bundle before you trust it with the whole machine.
Felo SuperAgent on dsh
DeepSeek Harness
DeepSeek HarnessWork that outlives the session that started it
A dsh session ends when you close it. Some jobs should not. SuperAgent runs the task on a persistent canvas, keeps what it learned, and stops when it genuinely needs you.
Mount it
cp -r felo-skills/felo-superagent .dsh/skills/Felo SuperAgent on dsh
- Bundlefelo-skills/felo-superagent
- Runs onA persistent canvas
- Interrupts youOnly when blocked
- KeepsState between runs
This skill runs on
Read the DeepSeek Harness guideDeepSeek Harness
Open-source agent harness (developer preview)
dsh reads skills straight out of folders you point it at, so a Felo bundle is one copy away from being callable in a session — and one delete away from being gone.
- Loads from
- ~/project/.dsh/skills/ · ~/.agents/skills/
- Install is
- A folder copy, no build step
- Worth knowing
- Only direct children of a root are read
A job with states you can inspect
Long-running work is only trustworthy if you can see where it is. Each stage records what it did and what it is waiting on, so you can leave and come back without re-reading a wall of output.
- Queued
Scope agreed
The job records the question, the sources it may use, and what counts as done.
- Running
Collecting
Pulls from search, pages, and documents, writing each finding to the canvas as it lands.
- Blocked
Needs a decision
Two sources disagree on a number. It stops and asks rather than picking one silently.
- Running
Drafting
Writes the deliverable against the agreed format, keeping the source list attached.
- Done
Ready to review
Hands back the draft, the sources, and the open questions it could not close.
Add it on dshMount it where it can survive a session restart
Everything else on this page is a tool the session calls. This one is closer to a job the session hands off, so the mount point matters more than usual.
~/.agents/skills/UserLoads in every project, and is the same folder other harnesses on this machine look at.
Mount the bundle
mkdir -p ~/.agents/skills
cp -r felo-skills/felo-superagent ~/.agents/skills/Prove it loaded
List the skills you can call from this project.Confirming a run is genuinely persistent
The failure mode here is a run that looks continuous but is actually starting over each time.
Start work, then close the session
Reopen and ask what the job has done so far. A real canvas answers with detail you did not paste in; a fake one starts the task again.
Trigger the blocked state on purpose
Find a question where two sources disagree. The run should stop and ask. If it picks a side without telling you, the state machine is decorative.
Check the source list survived
Ask for the sources gathered two stages ago. Persistence that loses provenance is worse than no persistence.
Jobs worth handing over rather than running
Use it for work measured in days, or work that has to keep watching after you stop.
Track this vendor's pricing and packaging pages. Re-check weekly, report only what changed, and never contact anyone.What comes backA standing brief that arrives as a diff against last week rather than a fresh summary, so a quiet week is visibly quiet.
Keep researching this market while I work on other things. Stop and ask me whenever two sources disagree on a number.What comes backA background question that progresses between sessions and only interrupts you for real decisions.
Feed the run, then publish what it found
A long-running job needs inputs it can trust and somewhere for the output to go.
What persistence does not buy you
A canvas is memory, not judgement. Two things stay firmly on your side of the line.
It will not know when to stop
A recurring job with no end date keeps running. Set the cadence and the stopping condition when you start it, not when you notice the bill.
It will not notice the world moved on
If the underlying question changes — a competitor pivots, a policy shifts — the run keeps answering the old version until you tell it otherwise.
It is not a scheduler you should forget
Check the first two or three outputs closely. A run that starts slightly wrong compounds, and the cost of catching it early is much lower.
What dsh users ask about long-running work
Does the run keep going when my machine is off?
The job lives on the Felo canvas, not in the session, so closing the harness does not end it. Reopening the session picks the thread back up where it stopped.
How is this different from just re-running the prompt?
Re-running starts from zero, including the questions you already answered. The canvas keeps those answers, so the second and fifth runs spend their effort on the parts that are still open.
Can I stop a run that is going wrong?
Yes, and it is worth doing early. Because each stage writes to the canvas, you can see which stage introduced a bad assumption instead of discarding the whole run.
Every Felo skill for DeepSeek Harness
Five skills written for DeepSeek Harness. They follow the same pattern on this platform, so you can combine them in one session without re-explaining your setup.
Felo Search
Felo Slides
Felo YouTube Subtitling
Felo X SearchOne API key. A full agent toolkit.
Start with a skill, then compose the exact models, live data, knowledge, and output tools your dsh workflow needs — the same key every Felo skill already uses.
Models
Use cost-efficient LLM access when your agent needs custom reasoning or generation.
- LLM API
- Chat completions
- Model access
Live information
Give agents current web and social signals instead of relying on stale context.
- Web Search
- X Search
- Web Fetch
Knowledge and output
Ground work in reusable knowledge, then turn it into shareable deliverables.
- LiveDoc
- PPT generation
- Mindmaps
Creative tools
Extend the workflow from research to visual and office-ready results.
- Image generation
- Page tools
- Office outputs