agentleFS
CRELLIA-S-LCRELLIA-S-L/srs-dd/.claude/skills/srs/SKILL.md

srs

Skill0 starsChanged 32 days agoMITView on GitHub

Working with the project specification (SRS in specs/). Invoke ALWAYS before any code change that alters behavior; when planning any task — including planning a feature, breaking a task into steps, or sequencing work across several requirements; when asked what the system should do; when investigating why something is built the way it is; when changing or cancelling a requirement. For a guided dialog that authors a brand-new requirement, prefer the srs-new skill. Not needed only for changes that do not touch behavior — typos, formatting, comments.

2 other projects use a version of this.

Adopted by

Most recently taken up by

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:

    Read what the project says of itself: specs/00-glossary.md, specs/01-introduction.md, specs/02-overview.md — short, and in its own words.

    line 15

≈ “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
<!-- Copyright (c) 2026 CRELLIA S.L. — part of SRS-DD, distributed under the MIT License -->

Working with the specification

The rules of the format are in srs-dd/standard.md, by section, and are not restated here. python3 srs-dd/tools/srs_view.py --vocabulary prints the words a block may use — statuses, fields, areas, modal verbs; open the standard for a rule, not for a word. Statements, titles and rationales are written in the lexicon of specs/srs-config.json, whatever language this file is in.

A question about the system

  1. Read what the project says of itself: specs/00-glossary.md, specs/01-introduction.md, specs/02-overview.md — short, and in its own words. A lookup by number needs a number nobody has handed you; a search finds the word you guessed.
  2. Narrow: python3 srs-dd/tools/srs_view.py --areas, then python3 srs-dd/tools/srs_view.py --list --area <AREA>, then python3 srs-dd/tools/srs_view.py <ID>, then --code <path>.
  3. Where the specification does not say, name the rungs you walked — the area, the words, the paths. Where the question is undecided, specs/91-open-issues.md holds it; read an entry whole.

Before changing code

  1. python3 srs-dd/tools/srs_view.py --code <path> --statements — the requirements behind the file, from the code and tests fields and, for a file, its own annotations, each with what it obliges; a directory works. Without the tool: grep -rn "<path>" specs/ and the table in specs/90-traceability.md, whose Incoming links section is every link in one place.
  2. Read what the answer names: python3 srs-dd/tools/srs_view.py <ID> --where prints every annotated line that carries the requirement, and python3 srs-dd/tools/srs_view.py <ID> --where --source prints the region each marks; where its header says no srs-end: ended it, read on in the file. Read those, in one call, before any file whole. A file the fields name that no line speaks for is printed as such, and is the one to open. A search over the repository is the fallback, taken out loud (AGENTS.md).
  3. Found some: choose by the statements the ones the change will touch, and read those in full — python3 srs-dd/tools/srs_view.py <ID> resolves the links both ways — with what each derives_from and depends_on. Incoming links are the blast radius; the rest of the list is known by its statements and is not opened.
  4. Found nothing: the behaviour is undescribed. Author a requirement first (srs-new); never code silently.
  5. With a grounds/ directory beside specs/: python3 srs-dd/tools/srs_grounds.py --blast <path> — what the requirements stand on. refuted or expired is not a stop; it is said, because the change inherits that ground. Procedure: srs-bet.
  6. With an arch/ directory: grep -n <path> arch/90-map.md, or the directory above it, since a directory in carries owns what is under it — the row names the element the file belongs to, what else it carries and the requirements it holds. A change that moves a responsibility between parts is said, not left for the checker. Procedure: srs-arch.
  7. Report the lookup, not only its outcome — the paths given, the words searched. What you say about a file is read off the file in that report and re-read before sending; that re-read is also where a figure of speech standing in for a plain verb comes out, and where a clause the reader cannot act on is deleted. Connected prose; a list only for what the reader counts. Every record named to a person is cited at its first mention in the message, from python3 srs-dd/tools/srs_view.py --cite <ID>….

Two acts

Authoring ends at the written requirement; building it is a separate act, started deliberately, often by somebody else. This procedure is the second act; srs-new is the first. Sliding from one into the other leaves a specification that records only what shipped.

Building a requirement

In from a file (above), or from an approved requirement nobody built: python3 srs-dd/tools/srs_view.py --coverage, grep -rn "status: deferred" specs/; read it with its links both ways.

  1. The requirement still says what you are about to build. If not, you are authoring — stop, srs-new.
  2. The plan names requirements, not work — "fix the storage layer" is not a step — each cited at first mention from --cite; constitution articles ART-* by number. The plan may not conflict with the method's articles or, where the project keeps one, specs/constitution.md.
  3. Code.
  4. Name what you chose: where the requirement could have been met another way, the way taken goes to specs/adr/ (Workflow, srs-dd/standard.md), or the report says none.
  5. Close the loop, in the same set of edits: re-read every named statement against what was built; what it does not describe is written in or taken out. A reworded statement passes the judgement srs-new gives a new one, against the qualities srs-dd/standard.md requires of a statement, and the same lookup — what else speaks to it, what points at it, the FR-DOC-* sections and the files in its code that restate it — and the finding is said before the edit. Then status per Lifecycle, code and tests with real paths, and every file named there says so back with implements: / verifies: above what carries it out, ended by srs-end: (Annotations, srs-dd/standard.md): the field is the specification's claim, the annotation is the file's. An open issue the change settles leaves specs/91-open-issues.md.
  6. python3 srs-dd/tools/srs_check.py, then the srs-check procedure: suites offered, inspections named.

A fix that adds what the statement is silent about means the statement is wrong. A change naming no requirement means no behaviour moved or a requirement is missing.

Withdrawing a requirement

withdrawn cancels with no successor; with one, it is superseded and the successor is named. Before the status changes:

  1. python3 srs-dd/tools/srs_view.py <ID> — the incoming links, grouped by field: depends_on loses its meaning, derives_from its reason, refines the rule it sharpened, conflicts_with nothing.
  2. Show the direct dependants in full, count what lies beyond; settle one level per decision — each resolution acts on the direct dependants, and may be a withdrawal with a tree of its own.
  3. Settle each dependant with the maintainer — none is a default: do not withdraw; narrow; supersede; cascade, each dependant back through this section; re-parent; promote and reword; orphan deliberately, and say so; stage it — intent, dependants, withdrawal last.
  4. Set the status and say why in the rationale, its only home. The number stays dead.

The checker flags what rests on a withdrawal — a guard, not a substitute.

Planning multi-requirement work

  1. Resolve the requirements in scope and their closure: depends_on, derives_from, refines, and the incoming links.
  2. Behaviour with no covering requirement: the first steps author it (srs-new).
  3. Order the steps by depends_on; each step cites its requirements (--cite), the articles that bind it ("per SRS-DD-ART-030") and the verification method.
  4. A draft in scope is a blocker, marked so (SRS-DD-ART-020). A superseded one sends the plan to its successor; a withdrawn one is an error to surface.
  5. The last two steps, always: close the loop; the checker and the srs-check procedure. A step that builds or tests is marked "requires the user's explicit confirmation each time" (SRS-DD-ART-030).

The plan lives in the conversation, never in specs/; a plan is not approval — statuses are (SRS-DD-ART-020).

Also

  • Prohibitions: What not to do in srs-dd/standard.md; and no builds or tests without the user's explicit confirmation (SRS-DD-ART-030).
  • What is put to the maintainer for a decision — a statement, a rewording, a step of a plan — is in the message that asks, as it would be written, never "as above".
  • A mismatch between code and requirement: fix neither side; record it in specs/91-open-issues.md and tell the user. A finding ends in therefore — must be fixed; is deliberate, and why; needs the maintainer's decision, with the options. Without one it is an observation, and is worked out or dropped, not handed over.

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.