offload-task
Skill4 starsChanged 17 days agoMITView on GitHub
Write one stuck problem into a standalone markdown brief so a fresh session in any coding agent can attack it without inheriting the dead ends. Use when the user says they want to offload a problem, take it to a new chat or another agent, or asks for a prompt/doc explaining a problem they need help with.
In this repository17 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- ask-firstSkill
- claude-project-rulesSkill
- compact-markdownSkill
- humanizeSkill
- pr-reviewSkill
- empty-commitSkill
- github-issue-creatorSkill
- git-mergeSkill
- git-worktreeSkill
- handoff-sessionSkill
- html-landing-pageSkill
- readme-craftSkill
- skill-auditSkill
- study-notesSkill
- visualize-itSkill
- zoom-outSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 0 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 9
Not a session summary. handoff-session covers the whole session so work continues.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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
Extract ONE problem from this conversation into a markdown brief saved to ~/Downloads/.
Not a session summary. handoff-session covers the whole session so work continues; this covers one unsolved problem so someone else solves it. Everything unrelated to that problem is excluded, even work done this session.
Scope
- Arguments given → that is the problem. No arguments → the most recent unresolved problem.
- Unsure which of several → ask before writing.
Reader
A coding agent with this repo checked out — no prior context, full file access.
- Reference
path/to/file.ts:123. Do not paste code the reader can open. - Inline only what cannot be opened: error output, terminal logs, user observations of the live UI, values from screenshots.
- Name the project rules that constrain the answer (
.claude/rules/*.md,CLAUDE.md) by path — do not restate them.
Attempts are the payload
The reason to offload is that the current thread burned attempts. List every one, one line each: what was tried → what happened. This is the section that stops the new session re-proposing a failed fix, so never compress it away.
Include attempts made in code AND things the user checked by hand.
Facts, not theories
Report observations. Omit your own diagnoses, hunches, and "likely cause" reasoning — those are what the thread got stuck on, and repeating them anchors the new session onto the same track.
Unknowns belong in Open Questions as questions, not as answers.
Exception: a hypothesis already tested and disproved belongs in Attempts as a fact.
Structure
# [Problem in one line]
## Goal
What should happen, in the flow it belongs to.
## Failing behaviour
What happens instead. Verbatim error output. Exact file:line.
## Reproduce
Command, spec, or steps.
## Attempts
- Tried X → Y happened. Code attempts and manual checks both.
## Evidence
Observations from the live app, logs, screenshots, network/console.
## Open questions
Questions, unanswered.
## Constraints
Rule files that apply. Anything the new session must not do.Drop any heading with nothing real to put under it.
Output
- Save as
~/Downloads/[<Project>]-offload-<TICKET>.md, else[<Project>]-offload-<topic>.md. Hyphenated; the square brackets around the project name are literal. <Project>= repo name fromoriginremote URL, else folder name; cut to 20 characters.- Name exists → append a timestamp. Never overwrite.
~/Downloadsmissing or unwritable → ask for a destination.- Redact secrets, credentials, PII.
- Report the saved path. Do not print the document in chat.
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.

