capturing-decisions
Skill0 starsChanged 50 days agoMITView on GitHub
Use when a decision, correction, or design rationale emerges that would otherwise be lost when the session ends.
In this repository20 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- templates/CLAUDE.mdCLAUDE.md
- choosing-subagent-modelsSkill
- committing-changesSkill
- debugging-systematicallySkill
- delegating-workSkill
- designing-changesSkill
- exploring-a-codebaseSkill
- injecting-dependenciesSkill
- planning-changesSkill
- pruning-testsSkill
- reflecting-on-changesSkill
- resolving-conflictsSkill
- reviewing-codeSkill
- verifying-changesSkill
- writing-commentsSkill
- writing-logsSkill
- writing-project-instructionsSkill
- writing-skillsSkill
- writing-testsSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 5 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
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
- Distinctive: line 20
No decision home in the project: ask before creating one.
- Distinctive: line 23
Test: would someone cloning this repo need it?
- Distinctive: line 34
Do not record what the repo already says: code structure, git history, the project's docs.
- Distinctive: line 47
Before acting on a record that names a file, function or flag, verify it still exists.
- Distinctive: line 48
Delete records that turn out wrong.
≈ “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.
Files
Capturing decisions
Session state is disposable. Anything that must outlive the session is promoted into the repo, wherever this project keeps its decisions — at the moment it is decided, not at the end. There is no end: in the repo this kit came from, plans piled up for months while the decisions file beside them stayed an empty header.
| What | Goes to |
| A decision and its reasoning | The project's decision record — docs/adr/, docs/decisions/, a decisions.md. After /vvkit:onboard, .claude/context/decisions.md |
| A validated design | Where the project keeps designs. After onboarding, .claude/context/specs/ |
| A correction about how to work | Durable memory, when the harness has one |
| A plan in flight, scratch, tool state | The session's scratch area — .agents/ after onboarding — and stays there |
- No decision home in the project: ask before creating one. A new directory is a convention imposed on everyone who clones the repo.
- Promote into the repo it is about. A design for another project goes to that project's repo. Test: would someone cloning this repo need it?
What earns a record
Record it when the reasoning is not recoverable from the code:
- A choice between viable options, and why this one.
- A constraint not visible from any single file.
- A rejected approach and what ruled it out — the one most needed and least written.
- A correction from the human about how to work.
Do not record what the repo already says: code structure, git history, the project's docs. Asked to remember something already there, record what was non-obvious about it instead.
Shape
- One fact per entry: what was decided, why, and what it rules out.
- Absolute dates — "last week" is meaningless in six months.
- Link related records. A link to one not written yet marks something worth writing.
- Deferred work carries a date, a reason, and what brings it back — "when a second channel exists". An old deferral whose trigger never came has expired: delete it or schedule it.
Recalled records are stale by default
- Before acting on a record that names a file, function or flag, verify it still exists.
- Delete records that turn out wrong. An outdated record is not harmless history.
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
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.

