agentleFS
BoredHFBoredHF/albertskills/skills/intent-diff/SKILL.md

intent-diff

Skill2 starsChanged yesterdayMITView on GitHub

Compares what the user actually asked for with what you delivered, point by point, to catch drift, missed requirements and unrequested changes, then fixes or flags them. Use before declaring any multi-part task done, at the end of long sessions, after working from a pasted spec, or when the user says "did you do everything", "check against what I asked" or "intent diff".

The file

We split this skill into the rules it gives an agent

2 rules. About 0 are common, 2 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Check numbers and literals literally.

    line 22
  2. Distinctive:

    Do not re-litigate the request.

    line 26

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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

Intent Diff

Done means the request is met, not that the plan was finished. Diff the delivered work against the user's words, never against your own plan.

Scale

  • Single-ask task: skip.
  • 2 to 4 asks: check silently and print the one-line clean form only, or the exceptions.
  • Multi-part request, pasted spec or long session: full version.
  • This checks that the work matches the request. Whether the work is good is red-team.

Hard rules

  1. Re-read every user message of the task verbatim, not your summary. Later messages amend earlier ones.
  2. Quote the user's words for each requirement. Never paraphrase.
  3. Evidence or it is not done. Open the file, run the thing, count the items. "I believe I did" is Partial at best.
  4. Check numbers and literals literally. "3 options" means count three, "under 100 words" means count, names are spelled exactly.
  5. Unrequested work is drift by default: refactors, renames, extra files, added features, reformatting, "while I was there".
  6. A broken constraint ("don't touch X" and you touched X) is a Different requirement, not unrequested work. Undo it under Fixed now.
  7. A requirement you chose not to meet is said out loud, never silently dropped.
  8. Do not re-litigate the request. Do it as asked or flag it.

Process

  1. Re-read all user messages of the task.
  2. Split into atomic requirements: explicit asks, constraints ("only", "don't", limits, formats, names), and at most 3 implied ones, marked implied.
  3. For each, find evidence in the delivered work and set a status: Done, Partial, Missed or Different.
  4. Reverse diff: list everything delivered that maps to no requirement. Mark each Keep (name the requirement it serves), Revert or Flag.
  5. Fix now: finish Missed and Partial items. Correct Different items, or flag them with the reason. Revert unrequested work unless it is necessary for a requirement.
  6. Report exceptions only. Do not print Done rows.

Output format

INTENT DIFF: <n> requirements · <n> already done · <n> fixed now · <n> flagged

Fixed now:
- "<quote>" → was <missed|partial|different>; <what was done>

Flagged (your call):
- "<quote>" → <why it was not met or is Different, and the options>

Unrequested:
- <thing>: kept, serves "<quote>"
- <thing>: reverted
- <thing>: flagged, your call (<why it may be needed>)

Omit empty sections. If clean, print exactly one line:

Intent diff: 7/7 done, nothing unrequested.

Anti-patterns

  • Diffing against your plan instead of the request.
  • Printing the whole table of Done rows.
  • Marking Done from memory without opening the work.
  • Calling extras "improvements".
  • Flagging what you could simply fix.
  • Treating the latest message as the whole request.
  • Paraphrasing a requirement until it matches what you built.
  • Reverting an unrequested change without checking that a requirement does not depend on it.

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.