agentleFS
RomeCoreRomeCore/LLMDesktopAssistant/.agents/skills/dass-report/SKILL.md

dass-report

Skill1 starsChanged 32 days agoMITView on GitHub

File a structured report about dASS itself - a native tool that behaved unlike its description, a prompt, an API, or something only the user can see, like the UI or localization. Use when a tool does something other than the model expected, when the same friction is hit a second time, when the user reports a UI or localization problem, or when the user asks for an improvement to be written up.

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
  1. Very specific:

    Every report lands in docs/agent-reports/ as AR-NNNN-<slug>.md.

    line 10

≈ “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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

dASS report (dASS)

A tool's description is its contract, and the client is always right: when a tool's behaviour contradicts the plain reading of its description, the contract is what failed. A report is written against the contract, not against the model that misread it.

Every report lands in docs/agent-reports/ as AR-NNNN-<slug>.md. One file, one problem. Write for the maintainer, who was not in this session.

Two lanes

LaneWhere the friction comes fromSurfaces
agentI hit it myself, while workingnative tools and their descriptions, argument schemas, error texts, prompts I read, APIs (Lua, addon loading)
userOnly the user can see itUI, localization, visuals, settings screens

The lane decides who supplies the facts. In the user lane I cannot observe the thing at all, so I interview for what the report needs and record what I was told - the gaps get filled by the user or they stay gaps. In the agent lane the evidence is mine, and it is exact: what I called, and what came back.

Steps

1. Say it in one line, and offer

The moment friction happens, say it in one line and offer to file it. The offer is the filter: friction that does not survive one question is not worth a file.

In the user lane the user opens the topic, so step 1 is asking for the one-line version rather than offering one.

Done when: the friction is on the table in one line, and the user said yes - or asked for the report directly.

2. Look for the report that already exists

List docs/agent-reports/ and grep it for the tool name, the surface and the symptom. A second sighting of the same friction is the same report: append a witness and stop. Two files about one problem split the evidence and get read as two complaints instead of one pattern.

Done when: either an existing report grew a witness line, or the folder is known not to hold it.

3. Take the number

AR-NNNN - the next free number after the highest in the folder, taken from the listing made in step 2. Check the filename is free immediately before writing; another session may be writing at the same moment.

Done when: the filename exists in nobody's folder but mine.

4. Write it

The template below. Every section filled; an empty Unverified is fine, an empty Witnesses is not.

Digging through the dASS source to pin the cause is optional. This is a report, not a patch, and a session spent outside the dASS repository should not turn into a code dive over one rough edge. When a claim comes from the source rather than from behaviour, say so in the report and mark it as traced.

Done when: the file is on disk with a lane, a severity, a repro and at least one witness.

5. Show the path

The path plus two or three lines of summary. The user will open the report; retelling it wastes the retelling.

Done when: the user has the path.

The template

# AR-0001 - <the friction in one line>

- **Status:** open | fixed | wontfix
- **Severity:** silent | visible | friction
- **Lane:** agent | user
- **Surface:** tool `fs-edit` | prompt | API | UI | localization | docs
- **First seen:** YYYY-MM-DD

## What I expected

The behaviour the tool's own description promises, in the model's reading of it.

## What happened

What actually came back - the call, the output, the file that ended up on disk.

## Why it costs

What the surprise costs per occurrence: steps, tokens, attention, or the correctness of the result.

## Repro

The smallest sequence that shows it. A call the maintainer can repeat.

## Witnesses

- YYYY-MM-DD - <session> - <what I was doing when it happened>

## What I did instead

The workaround. A workaround that costs three calls is the argument for the fix.

## Proposed fix

Options, cheapest first. A traced cause goes here, marked as traced.

## Unverified

What was inferred rather than observed, and what could not be checked from where I stood.

Status changes: open → fixed when the behaviour is gone, wontfix when the user decides against it. A wontfix keeps its reason in the report - it is what stops the next session from re-filing the same thing.

Severity

LevelWhat it means
silentThe tool lied and the result looks right. Nobody notices until much later. The worst class.
visibleIt failed loudly. I saw it, and worked around it.
frictionIt works, but every occurrence costs steps, tokens or attention.

Not here

  • A fix I can write myself, as an addon → equip.
  • A feat worth remembering → record-achievement.
  • A design idea the user wants to pursue on their own terms → docs/ideas/.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.