Replace custom SVG artwork with emoji-based stone icons: update .codex-plugin/icon.svg and .codex-plugin/logo.svg to simple SVGs that render the 🪨 emoji (add role/aria-label and emoji font fallbacks). Also simplify README by inlining the stone emoji instead of referencing the SVG image. This reduces asset complexity and ensures consistent emoji rendering and accessible labels.
- Punchy, direct prose — no filler, no fluff
- Dynamic GitHub badges (stars, last commit, license)
- Before/After comparison table up front
- Problem framed as table for quick scanning
- All technical content preserved: phases, parallel execution,
Codex adversarial review, configuration, commands, methodology
- Skills listed in table format inside collapsible section
- Star history chart and ecosystem links
- Consistent with caveman README style
https://claude.ai/code/session_012D3nKU3keqDLHpk7ustfUc
- Rename all files/directories: .blueprint -> .cavekit, cmd/blueprint -> cmd/cavekit,
context/blueprints -> context/kits, skills/blueprint-writing -> skills/cavekit-writing,
agents/blueprint-reviewer -> agents/cavekit-reviewer, scripts/blueprint* -> scripts/cavekit*
- Update Go module path: github.com/julb/blueprint-monitor -> github.com/JuliusBrussee/cavekit
- Replace all content references from blueprint/Blueprint to cavekit/Cavekit
- Update plugin.json, install.sh, codex-plugin config, and all documentation
- Add caveman ecosystem link to README
- Binary renamed from blueprint to cavekit
Part of the Caveman ecosystem: https://github.com/JuliusBrussee/cavemanhttps://claude.ai/code/session_012D3nKU3keqDLHpk7ustfUc
Closes the gap where acceptance criteria specified in draft could be lost
during architect and build phases. Adds:
- Coverage Matrix to build-site format: maps every acceptance criterion to
its assigned task(s), making gaps visible before build starts
- Pre-flight coverage check in build: flags unassigned criteria before
executing any tasks
- Post-flight blueprint verification: cross-references completed tasks
against original blueprint criteria after build, adds remediation tasks
for anything missed
- Quick mode coverage gate: lightweight but mandatory criterion check
- Criterion-level validation rules in architect command and agent
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Introduce a first-class DESIGN.md design system and wiring across the pipeline. Adds a new /bp:design command (create/import/extract/audit/update DESIGN.md), a bp:design-system skill, and a design-reviewer agent to validate DESIGN.md. Update agents (drafter, architect, task-builder, inspector, surveyor, surveyor, etc.) and commands (draft, architect, build, inspect, quick, revise, init, help) to read and enforce DESIGN.md: reference tokens in blueprints, include Design Ref in tasks, report DESIGN VIOLATION statuses, and log design changes to context/designs/design-changelog.md. Create context/designs/ conventions and changelog handling; add UI-craft skill and several UI reference docs. Overall: adds design-system as a cross-cutting constraint layer so UI decisions are tokenized, referenced, and audited end-to-end.
Introduce a canonical bp-config.sh tool to manage Blueprint execution presets and resolved models (reasoning, execution, exploration). Add CLI, validation, preset resolution, summary, and listing functions plus a test script (scripts/test-bp-config.sh). Integrate bp-config into the toolchain: update docs (README, help) and add a new command spec (commands/config.md); update command docs (draft, architect, build, quick, inspect, research, gap-analysis, etc.) to resolve and use explicit models via bp-config.sh; add bp-config to allowed-tools where appropriate. Make codex-config.sh a backward-compatible wrapper that sources bp-config.sh, and have setup-build.sh source bp-config.sh and emit the preset summary. This centralizes model preset logic and makes model selection explicit for delegated subagents.
Add website blueprints and planning/implementation tracking for the new marketing site and docs.
- Added blueprints: context/blueprints/blueprint-landing-page.md and blueprint-docs-page.md (detailed requirements & acceptance criteria).
- Added site plan: context/plans/build-site.md (19 tasks across tiers) and moved several context/sites/* files into context/plans/ (renamed). Deleted old context/sites/build-site.md.
- New implementation tracking: context/impl/impl-website.md with task table for landing + docs; updated context/impl/impl-cli.md task list and timestamps.
- Commands/docs updates: commands/build.md (argument hint now includes [FILE] and adds scoped impl-file reading rule) and commands/inspect.md (add Restore Archived Impl Tracking instructions to recover archived impl-*.md files).
- Misc: updated context/blueprints/blueprint-overview.md metadata and domain table to reference the website blueprints; added ignores to .gitignore (/.playwright-mcp, current-site-full.png); modified scripts/setup-build.sh.
These changes prepare the repo for implementing a zero-build, two-file gh-pages site (index.html + docs.html) and ensure the build/inspect workflows handle scoped impl files and archived implementation tracking.
Add plugin packaging for the Blueprint Codex plugin: include privacy and terms documents, icon and logo SVGs, and a screenshots placeholder. Add .codexignore to exclude dev/build artifacts and context directories. Update .codex-plugin/plugin.json to register agents/commands directories and surface UI metadata (composerIcon, logo, screenshots, privacyPolicyURL, termsOfServiceURL). Clear TODO.md contents (remove prior task list). These changes prepare the plugin for distribution and compliance with marketplace metadata requirements.
Broadened file discovery from *site* only to *site*, *plan*, or *frontier*
(excluding *overview* index files) so /bp:build, /bp:progress, and
/bp:inspect can find buildable task graphs regardless of naming convention.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Update merge and cleanup guidance to require removing git worktrees before deleting branches. Docs now instruct merging subagent branches one at a time, running validation per-merge, then running `git worktree remove <path>` before `git branch -D <branch>`, and to use `--force` when cleaning up conflicted worktrees. Clarifies that Claude Code only auto-cleans worktrees when agents made no changes and that callers must clean up worktrees for committed changes to avoid branch-delete failures.
Implements the deep research design spec: a parallel multi-agent research
system that grounds blueprint design in real evidence (codebase analysis +
web search). Includes standalone /bp:research command with depth levels
and integration into /bp:draft flow with intelligent trigger and brief reuse.
- New: commands/research.md — standalone command with adaptive agent count,
two-wave dispatch (findings board coordination), Opus synthesizer, and
research brief output to context/refs/
- Updated: commands/draft.md — restructured step 3 (3a-3g) to insert
research check (3c), research phase (3d), and research-informed Q&A (3e)
- Updated: commands/help.md — added /bp:research to workflow and commands
- Updated: README.md — added Research phase to workflow diagram, commands
table, file structure, and methodology description
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Design for parallel multi-agent research system that grounds blueprint
design in real evidence. DAG-based orchestrator with adaptive 1-4
codebase agents + 1-4 web agents, shared findings board for coordination,
two-pass synthesis (synthesizer subagent + lead agent), named research
brief artifact in context/refs/. Hybrid command model: inline in
/bp:draft + standalone /bp:research.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Introduce context/plans as the primary location for build sites and add a new /bp:init command to bootstrap the context hierarchy. Add commands/init.md (idempotent, non-destructive scaffolding), update docs and CLI help to reference context/plans (with fallbacks to legacy context/sites and context/frontiers), and add post-build guidance to update per-directory CLAUDE.md mappings. Update site discovery (internal/site/discovery.go), several scripts (blueprint, blueprint-picker.ts, blueprint-status-poller.sh, dashboard-progress.sh, setup-build.sh) to search context/plans first, and adjust inspect/revise/progress docs to use the new convention. Also rewrite the context-architecture SKILL to reflect a DAG-based progressive disclosure model, and add /docs to .gitignore.
Design for organizing project knowledge as a DAG of documents with
progressive disclosure. Covers 4-tier information flow (refs → blueprints
→ plans → impl), full-repo CLAUDE.md hierarchy, /bp:init scaffolding
command, build-time CLAUDE.md updates, and backpropagation via /bp:revise.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add Codex integration and update build orchestration: • Add .codex-plugin/plugin.json and sync script to register Codex plugin and expose Codex-related skills. • Introduce multiple new blueprint specs (codex-bridge, command-gate, draft-challenge, speculative-review, tier-gate) and an implementation tracking file for Codex. • Update README to describe auto-parallel /bp:build behavior, remove separate /bp:parallel and /bp:merge flows, and clarify plugin syncing. • Add bp-quick command for end-to-end quick builds. • Improve architect output to include a Mermaid dependency/parallelization graph. • Strengthen builder and task-builder agents with explicit guidance to never falsely mark tasks DONE and to require tests/verification. • Remove commands/merge.md and commands/parallel.md. • Update scripts and tooling (including new scripts/sync-codex-plugin.sh and changes to various blueprint scripts) to support the new flows. Overall: integrate Codex review patterns and move the build loop to an auto-parallel model with clearer gating and review hooks.
T-305: bp_design_challenge_cycle runs up to 2 challenge-fix cycles,
outputs AWAITING_FIXES with structured critical findings for caller.
T-306: bp_draft_challenge_hook integrates between blueprint-reviewer
(Step 8) and user review gate (Step 9), sets BP_CHALLENGE_ADVISORY_OUTPUT
for user presentation and tracks challenge duration.
T-307: Graceful degradation already in bp_design_challenge (returns 2
when Codex unavailable), bp_draft_challenge_hook logs skip with timing.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
T-303: bp_design_challenge sends all blueprints to Codex with design-level
prompt, parses structured findings with categories and severity.
T-304: bp_collect_challenge_findings separates critical vs advisory,
bp_format_advisory_for_user generates user-facing markdown table,
bp_format_critical_for_fix outputs structured data for auto-fix loop.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add codex-design-challenge.sh with purpose-built prompt for adversarial
blueprint review (domain decomposition, coverage, ambiguity, scope,
assumptions) and structured finding parser with category/severity support.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- scripts/codex-gate.sh: Added bp_review_fix_cycle function that
orchestrates the review → evaluate → fix → re-review loop with
configurable max cycles (default 2). Returns 0 (clear), 1 (max
cycles exhausted), or 2 (awaiting fixes from caller).
- commands/build.md: Updated tier boundary check to use
bp_review_fix_cycle for the fix loop orchestration.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- scripts/codex-gate.sh: bp_tier_gate evaluates findings against
tier_gate_mode (severity/strict/permissive/off) and returns
blocked/proceed. bp_generate_fix_tasks outputs fix task descriptions
for blocking findings.
- commands/build.md: Updated tier boundary check with severity gating,
fix-task generation, re-review cycle, and max 2 cycles per tier.
- Fixed grep regex bug in codex-findings.sh and codex-gate.sh where
pipe character was interpreted as alternation.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
New slash command that invokes Codex adversarial review on demand.
Supports --base <ref> override, outputs findings in Blueprint format,
appends to impl-review-findings.md, and handles missing Codex gracefully.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
After all tasks in a tier complete, the build loop now checks
codex availability and tier_gate_mode config before running an
inline Codex review of the tier's cumulative diff. Gracefully
degrades when Codex is unavailable or tier gate is disabled.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- T-005: codex_available flag and nudge already in codex-detect.sh
- T-006: scripts/codex-review.sh — Codex adversarial review invocation with
diff computation, finding parser (P0-P3), source:codex tagging, and
graceful fallback when Codex unavailable
- T-007: scripts/codex-findings.sh — Finding management with extended format
(Source, Tier columns), append/update/blocking-list functions, old-format
migration support
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Emphasize that committing is required before finishing to avoid losing uncommitted work: added a MUST-commit note in agents/task-builder.md and marked the commit step as CRITICAL in commands/parallel.md. Clarify subagent merge behavior: worktree branches are cleaned up automatically but branch refs and commits persist, include how to fetch and merge subagent branches, advise deleting local branch after merge, and skip merging when Agent reports "no changes were made."
Introduce parallel task execution support: add detailed README sections describing `/bp:parallel` in-session parallelism and `blueprint monitor` multi-session mode, and update CLI command names in the reference. Add a new task-builder agent spec (agents/task-builder.md) that defines the single-task subagent workflow for isolated worktrees. Add a bp-parallel command implementation doc (commands/parallel.md) describing orchestration, dispatch, merging, circuit breakers and critical rules. Implement frontier computation (internal/site/frontier.go) and comprehensive unit tests (internal/site/frontier_test.go) to determine ready tasks and render human-readable summaries.
Addresses 5 blocking + 6 minor issues from spec review:
- Increase text-secondary alpha for WCAG AA contrast
- Split blueprint-dim into text vs decoration tokens
- Make red-diagnostic fully opaque
- Pin deployment to gh-pages branch
- Specify concrete mobile breakpoint layouts
- Add footer section, enumerate section annotations
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Julius Brussee