BuildCLI-Agents
CLAUDE.md3 starsChanged 11 days agoMITView on GitHub
Pipeline skills — invoke by name to run a stage: - survey → profile the repo into .buildcli/context.md, split into bands - brief → requirements with testable acceptance criteria (moves .buildcli/active) - shape → architecture and a phased plan - worklist → atomic units with dependencies and band tags - build → execute the worklist band by band, with progress tracking - audit → check the implementation against the brief's acceptance criteria - patch → minimal defect fix (add --trace…
We split this CLAUDE.md into the rules it gives an agent
2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
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
- Very specific: line 30
Run the survey skill first to create or refresh .buildcli/context.md.
- Distinctive: line 32
Before any task, load only the band skill matching the work: service, interface, store, verify, delivery.
≈ “Common” is an estimate: at least 1 in 5 of the 25,680 CLAUDE.md 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
Claude Instructions
<!-- buildcli:autoload:start -->Slash commands: load command files from .claude/commands/*.md
Pipeline skills — invoke by name to run a stage:
- survey → profile the repo into .buildcli/context.md, split into bands
- brief → requirements with testable acceptance criteria (moves .buildcli/active)
- shape → architecture and a phased plan
- worklist → atomic units with dependencies and band tags
- build → execute the worklist band by band, with progress tracking
- audit → check the implementation against the brief's acceptance criteria
- patch → minimal defect fix (add --trace to file a defect blueprint)
- pulse → read-only pipeline snapshot: stage, unit counts, next step
- focus → move the active blueprint pointer without re-running brief
- forge → create or refresh a project-specific skill
- rig → configure Claude Code hooks and permissions
Band skills — load only the one matching the work in front of you:
- .claude/skills/service/SKILL.md
- .claude/skills/interface/SKILL.md
- .claude/skills/store/SKILL.md
- .claude/skills/verify/SKILL.md
- .claude/skills/delivery/SKILL.md
Also available:
- .claude/skills/design-review/SKILL.md → architecture tradeoffs before implementation
Startup behavior (required):
- Run the
surveyskill first to create or refresh.buildcli/context.md. - Refresh it whenever the stack, architecture, integrations, or standards change.
- Before any task, load only the band skill matching the work: service, interface, store, verify, delivery.
- Each band skill names the exact block of
.buildcli/context.mdto read. Read that block, nothing else. - Missing critical information → mark
NEEDS CLARIFICATIONand continue on safe defaults.
Runtime (always prefer it over reading state files by hand):
.buildcli/runtime/bcx band <name>— load exactly one context band.buildcli/runtime/bcx header— the shared header, without any band.buildcli/runtime/bcx active [path]— read or move the active blueprint pointer.buildcli/runtime/bcx next— units ready to start, grouped by band.buildcli/runtime/bcx graph— dependency graph, critical path, cycle report.buildcli/runtime/bcx claim|done|block <id>— unit state transitions.buildcli/runtime/bcx verify— run the project's test command.buildcli/runtime/bcx status --json— pipeline snapshot.buildcli/runtime/bcx resume— where the project left off, in one screen.buildcli/runtime/bcx doctor— validate context, graph, and configuration
Invoke it by that path, not as a bare bcx — it is project-local and not on PATH.
With rig --enforce applied, reading .buildcli/context.md directly is blocked
by a PreToolUse hook.
Shared state:
.buildcli/context.md— project context, split into [band:*] blocks.buildcli/active— path to the active blueprint directoryblueprints/<kind>/<slug>/— brief.md, shape.md, worklist.md, audit.md
Multi-agent relay:
- brief + shape → an analysis-focused agent (Gemini, Claude)
- build → a code-generation agent (Claude, Codex)
- Handoff happens through the files. The relay block at the end of every brief.md says who picks up next.
Bootstrap command:
./buildcli/scripts/bootstrap.sh --repo . --agent all --mode copy
<!-- buildcli:autoload:end -->In this repository19 other files it gives its agents.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
Reviews can't be read right now.
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.

