agentleFS
xmtpxmtp/libxmtp/.agents/skills/reviewing-specs/SKILL.md

reviewing-specs

Skill127 starsChanged 12 days agoMITView on GitHub

Use when reviewing a draft or amended spec in docs/specs/ before an owner sees it - covers the format checks, the admission test, precision, cross-spec contradictions, and sampling requirements against the code

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 3 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read docs/specs/SPEC-spec-format.md.

    line 8
  2. Distinctive:

    Prose: Every section's prose is a few direct sentences: what the mechanism is and what breaks without it, in the protocol's own terms and …

    line 16
  3. Distinctive:

    Contradictions: Read the related specs the Scope table names.

    line 18
  4. Very specific:

    Pending references: Every ?PREFIX marker names a registered prefix and says in prose what the missing obligation should require.

    line 19
  5. Distinctive:

    Legacy: Every obligation in the legacy source needs a disposition: carried, retired, or moved to a design note.

    line 22
  6. Very specific:

    Do not edit the spec. Do not set status: approved.

    line 29

≈ “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

Reviewing a spec

Read docs/specs/SPEC-spec-format.md. Review with a clean context, on a different model than the author. Your job is to find what is wrong, not to agree.

Checks, in order

  1. dev/nix-shell 'just spec-check' passes. Report what it warns about; the checker catches form, not meaning.
  2. Format. Section order; one obligation per row; condition then actor then MUST or MUST NOT; a two-to-seven-word title; at most three sentences; a Why cell that is empty or says what breaks, never a restatement; SHOULD only for an app or an operator; no SHALL; no file, module, or crate names; no phases; no platform choices.
  3. Admission. For each requirement, ask the six questions in SPEC §2. Name every requirement that fails, and which question it fails. Watch for the common failures: a rule whose violation only the violator observes, an internal data layout, a restatement of what the code happens to do, a requirement about tests, and one example of an invariant the spec already states.
  4. Precision. Every condition names a field, a stored value, or a comparison; every bound has a value or a rule; every fixed algorithm is named; every standard is cited by section with a link (SPEC-089, SPEC-074). List each requirement that uses a stand-in word (earliest, latest, confirmed, usable, bounded, sufficient) and say what it should name instead.
  5. Prose. Every section's prose is a few direct sentences: what the mechanism is and what breaks without it, in the protocol's own terms and without metaphor (SPEC-010). Flag prose that argues for the design, lists alternatives, or narrates decisions.
  6. Duplication. The obligation is not already stated in another spec (just spec-index), and not twice in this one: a rule and its value split into two rows, or two rows that a single act would violate, are one requirement (SPEC-028). An exact value appears in exactly one spec.
  7. Contradictions. Read the related specs the Scope table names. List any pair of requirements that cannot both hold.
  8. Pending references. Every ?PREFIX marker names a registered prefix and says in prose what the missing obligation should require. A marker naming an approved spec is a blocker: that obligation exists now and the reference must point at it.
  9. Reality. Sample at least ten requirements, weighted toward security, ordering, and error cases, and check each against the code anchors. A requirement the code does not satisfy is a waiver candidate, not a reason to weaken the requirement. Say which ten you sampled.
  10. Type blocks. A protobuf block must match the schema field for field, including field numbers, since those are normative. A WebIDL block must describe something that never reaches the wire. A foreign format such as an MLS object must be cited by document and section, not reproduced.
  11. Legacy. Every obligation in the legacy source needs a disposition: carried, retired, or moved to a design note. Challenge any retirement that drops an exact public value, a security limit, or an error contract.
  12. Unowned obligations. Definitions in the glossary and a spec's Terms name things; they must not carry obligations. An obligation stated only in a table other than a requirements table, a type block, or a definition has no owner, and no requirement to amend or evidence.

Report

A table: ID or section, severity (blocker, major, minor), the finding, and the suggested fix. Then the requirements you sampled in step 9. Distinguish what is wrong from what you could not verify.

Do not edit the spec. Do not set status: approved. If the spec is sound, say so briefly rather than inventing findings.

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.