DocSlime is a small CLI plus a skill pack for creating, filling, reviewing, and publishing an
opinionated docs/ tree. The name is silly on purpose; the method is not. It pulls the
scattered parts of a project into one integrated body, then gives future humans and AI agents
better context for the next change.
Why It Sticks
strategy -> product/design -> discovery -> requirements -> architecture -> testing
^ |
+----------- observation <- publishing <- verified build <------+
- Product and design context flows into
PRODUCT.mdandDESIGN.md, so agents stop
guessing what the code is for. - Continuous discovery lives in
experience/, where evidence and journeys become
solution-neutral requirements rather than an untraceable feature backlog. - Requirements and behavior flow into
REQUIREMENTS.mdandengineering/TESTING.md, so
TDD+BDD work can trace back to evidence. - Domain shape and tradeoffs flow into
engineering/ARCHITECTURE.mdandengineering/adrs/, so
Domain Driven Design language stays close to decisions. - Human judgment flows through
docslime-fill,docslime-adr, anddocslime-kiss, so
docs get filled, decisions get recorded, and bloat gets cut. - Delivery and observation close the loop:
engineering/PUBLISHING.mddefines promotion,
verification, and rollback;engineering/OBSERVABILITY.mdconnects production health and
user outcomes back to discovery.
Install DocSlime
Homebrew is the recommended local install path:
brew install DecisionNerd/tap/docslime
docslime init creates missing docs and reports what was created or skipped. It will not
overwrite existing files unless --force is explicit. The slime eats context, not your
worktree.
Other install paths stay available when Homebrew is not the right fit.
curl -LsSf \
https://github.com/DecisionNerd/DocSlime/releases/latest/download/docslime-installer.sh \
| sh
cargo install \
--git https://github.com/DecisionNerd/DocSlime \
--bins
First Run Path
docslime init
Create the standard docs tree and leave clear next steps in the repo.
npx skills add DecisionNerd/DocSlime
Use docslime-fill to interview the team and replace scaffold guidance with real context.
docslime add adr choose-storage-boundary
Record significant product and technical choices while they are still fresh.
Run docslime-kiss as an agent skill to find bloat, contradictions, stale placeholders, and
weak traceability before docs become ceremony.
The recommended happy path is short: install the CLI, run docslime init, add the skill
pack, fill docs/PRODUCT.md, then use docslime-kiss once the first useful context exists.
What DocSlime Creates
docs/
|-- README.md
|-- PRODUCT.md
|-- DESIGN.md
|-- REQUIREMENTS.md
|-- strategy/README.md
|-- experience/README.md
`-- engineering/
|-- README.md
|-- ARCHITECTURE.md
|-- TESTING.md
|-- PUBLISHING.md
|-- OBSERVABILITY.md
`-- adrs/README.md
docs/PRODUCT.md and docs/DESIGN.md are deliberately discoverable from the docs tree so
tools like impeccable can load product and design context without duplicate root files.
- Product + design:
PRODUCT.md,DESIGN.md, andstrategy/capture purpose,
voice, principles, and success measures. - Experience + requirements:
experience/captures evidence, opportunities, journeys,
and hypotheses;REQUIREMENTS.mdtranslates them into a testable build contract. - Architecture + ADRs:
engineering/ARCHITECTURE.mdandengineering/adrs/keep domain
boundaries and decisions explicit. - Testing:
engineering/TESTING.mdties TDD and BDD coverage back to requirements and journeys. - Publishing + observability:
engineering/PUBLISHING.mdand
engineering/OBSERVABILITY.mdcarry verified artifacts to
users and feed production evidence back into discovery.
Agent Skills
DocSlime keeps judgment-heavy work in skills instead of pretending every review belongs in a
CLI subcommand.
docslime-installanddocslime-initverify the CLI and scaffold the docs tree without
overwriting existing work.docslime-fillinterviews the user, replaces scaffold guidance, and keeps facts anchored
in the repo.docslime-adrcreates the next-numbered ADR and writes the decision in the project
vocabulary.docslime-kissreviews for contradictions, generic AI prose, weak traceability, and
overgrown docs.