operator-backlog-newitem
Skill1 starsChanged 4 months agoMITView on GitHub
Turn a request into a tracked backlog item instead of an interrupt. Use when the user asks for a feature, reports a defect, or describes work to do later - and whenever you find a defect mid-task that you are not currently authorised to fix.
In this repository23 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- .github/copilot-instructions.mdCopilot instructions
- templates/copilot-instructions.mdCopilot instructions
- speckit-analyzeSkill
- speckit-checklistSkill
- speckit-clarifySkill
- speckit-constitutionSkill
- speckit-convergeSkill
- speckit-implementSkill
- speckit-planSkill
- speckit-specifySkill
- speckit-tasksSkill
- speckit-taskstoissuesSkill
- backlogSkill
- code-intelligenceSkill
- field-notesSkill
- operator-backlog-refinementSkill
- operator-backlog-scrumSkill
- peer-agentsSkill
- request-auditSkill
- spec-drivenSkill
- worktreesSkill
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 80
Use it for what genuinely blocks the work in hand.
≈ “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
File a backlog item
$ARGUMENTS is the request, in the user's words if you have them.
A request that arrives mid-task has two possible fates: it derails the task, or it is held in your context. Context is the thing that does not survive — this machine's trace records 940 session exits and not one of them wrote a handoff, so everything held in context on those 940 occasions was lost. This command is the third option.
When to file
Always offer. File unprompted only when the request is unambiguously scoped work.
The error here is asymmetric and it is worth being explicit about which way. Filing a stray remark costs a human one rejection; failing to file loses the request entirely, which is the exact failure this command exists to stop. But an agent that files every utterance produces a backlog nobody reads, and an unread backlog is worse than none because it costs a reader time to discover that. So:
- File it, then say you did: "add X", "we should do Y", "Z is broken" — anything with a verb and an object you could hand to another agent.
- Offer, and wait: musings, half-formed preferences, anything where you would have to invent the scope. "I wonder whether..." is not a work item.
- Never file: things the user is telling you to do now. Do those.
Filing one
The tool writes the file. Do not hand-roll front matter and do not pick the id yourself — a second writer of this format is how the format drifts.
backlog new --title "One line naming the defect or the change" \
--evidence "What was observed, when, and how it reproduces" \
--why "Why it matters" \
--notes "Design questions, constraints, links"New items are filed proposed, meaning filed, not approved. That is the
default and you should leave it alone: filing is not approving, and an agent
that files an item as open has approved its own work.
Evidence is required and the tool refuses without it. Evidence is what was
observed, when, and how to see it again. "The mail system is unreliable" is not
evidence; "on 2026-08-05 operator inbox returned empty for a mailbox that
ls showed held two files" is. Quote the user verbatim where the request is
the evidence — a first-hand report in their own words outlives your summary of
it.
--spec defaults to none. Set it to a path under specs/ when the item
changes a specification, and add a requirement: line to the file afterwards
when you can name the exact requirement text. That is what makes the mapping
load-bearing rather than decorative: rename the requirement and the item goes
red, so somebody has to look at it again.
The one case where you may work what you filed
You are working an approved item and you find a defect. You may not work an unapproved item, and stopping to ask costs the session. The lawful move is to file the defect naming the approved item it blocks:
backlog new --title "..." --evidence "..." --blocks 12backlog ready will then list it, because item 12 is approved and this item is
what stands in its way. The exception is deliberately narrow — the blocked item
must itself be approved, so a chain cannot begin at something nobody approved —
and, more importantly, it is recorded. An agent with no lawful move does not
stop; it approves its own item and carries on, which repeals the gate while
appearing to honour it and leaves nothing behind that says so.
Use it for what genuinely blocks the work in hand. A defect you merely noticed
is a proposed item with no blocks, and it waits.
After filing
backlog check— the item you just wrote must conform.backlog newruns this for you and tells you when it does not.- Commit it. An item that exists only in your working tree is in exactly the same danger as one that existed only in your context.
- Tell the user in one line: the id, the title, and that it awaits their approval. Then go back to what you were doing.
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.

