agentleFS
anivaranivar/agentic-sdlc/skills/decision-records/SKILL.md

decision-records

Skill0 starsChanged 37 days agoMITView on GitHub

Keep a project's shared knowledge where the second reader can reach it: a glossary with one term per concept and the words to avoid, decision records for choices that are hard to reverse and surprising without context, dated research documents that record what a source taught, all as tables with honest calibration, naming systems and roles but never people. Triggers on: "ADR", "decision record", "glossary", "define the term", "what do we call", "record the decision", "research doc", "write down what we learned", "where should this live", a term used in conflict with the glossary, output that contradicts an existing record.

9 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

5 rules. 0 are common, 3 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 8 edited versions we read keep these.

  • 3

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

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

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

    Create files lazily: a glossary when the first term is resolved, docs/adr/ when the first decision qualifies, a research document when a …

    line 32
  2. Very specific:

    Before exploring an area, read GLOSSARY.md and the records under docs/adr/ that touch it.

    line 48
  3. Distinctive:

    When a term conflicts with the glossary, say so at once: "The glossary defines cancellation as X.

    line 50
  4. Distinctive:

    When a term is vague or overloaded, propose the canonical one.

    line 52
  5. Distinctive:

    When a statement about how something works disagrees with the code, surface the contradiction.

    line 54

“Common” is measured: kept by at least 6 in 10 of the 8 edited versions of this file we could read. “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

Decision records

Three kinds of record, each with a format in references/:

RecordLives atFormat
GlossaryGLOSSARY.md at the root, or one per context listed in GLOSSARY-MAP.mdreferences/glossary-format.md
Decision recorddocs/adr/NNNN-slug.mdreferences/adr-format.md
Research documentdocs/research/<date>-<question>.md, or the path docs/agents/conventions.md namesreferences/research-document.md

Create files lazily: a glossary when the first term is resolved, docs/adr/ when the first decision qualifies, a research document when a source taught something.

The rules

IdRuleIn practice
DECIDE-1A record for a hard-to-reverse, surprising trade-offAll three must hold. Easy to reverse: you will just reverse it. Not surprising: nobody will wonder why. No real alternative: nothing to record. One paragraph is enough.
DECIDE-2Glossary terms are opinionated and implementation-freeOne term per concept, the others under "avoid", one or two sentences on what it is. No implementation detail, no scratch notes, no specifications.
DECIDE-3Knowledge in tables with calibrationGlossaries, research documents and plan tables have fixed columns, in the repository, and every row's calibration is written (EVID-1): "unverified", never blank. The decision record is the prose exception, one paragraph. Prose elsewhere is for the reader who already has the table.
DECIDE-4Records carry no namesSystems and roles, never individuals.
DECIDE-5One file per topic, never another's fileCheck the path before writing. A file not created in this session is not overwritten; a new dated document sits beside the earlier one.
DECIDE-6Conflicts with a record are surfaced"Contradicts record 0003 (attribution patterns by default); worth reopening because …" rather than a silent override.

During design

  • Before exploring an area, read GLOSSARY.md and the records under docs/adr/ that touch it; the triggers for conflicts fire only on what was read.
  • When a term conflicts with the glossary, say so at once: "The glossary defines cancellation as X; this means Y. Which is it?"
  • When a term is vague or overloaded, propose the canonical one.
  • Stress-test relationships with concrete scenarios that probe the edges.
  • When a statement about how something works disagrees with the code, surface the contradiction.
  • Resolve a term, write it down then. Do not batch glossary updates.

What qualifies for a decision record

Architectural shape. Integration patterns between contexts. Technology choices with lock-in that would take a quarter to swap. Boundary and scope decisions, the explicit no as much as the yes. Deliberate deviations from the obvious path, so the next engineer does not "fix" them. Constraints not visible in the code. Rejected alternatives when the rejection is non-obvious.

Companion skills

NeedSkill
The questions that surface the decisionsgrill-plan
Calibration tags for the rowsevidence
The plan's Decisions table for reversible choicesplan-project

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.