Claude/init project 7l gu i#1
Merged
Merged
Conversation
Add the required claude.ai/code header, consolidate build commands, and document the template system, stack layering, practices pipeline, audit scoring, and /forge dispatch architecture. https://claude.ai/code/session_01JV7FJLEKixpyEzkRcru4Pj
- Scoring formula: recomendados now weight 50% real (perfect obligatorios = 7.0 not 10.0) - Swift lint: template uses swiftlint (consistent with swift-swiftui stack) - Stack detection: centralized in stacks/detect.md (was duplicated in 4 skills) - Bootstrap: creates .claude/agent-memory/ for agents with memory: project - implementer: removed reference to nonexistent specs/in-progress/ - README: corrected security checklist count 51→31 - forge.md: watch/scout marked as stubs - Git tags v0.1.0-v1.2.1 created (enables /forge diff) https://claude.ai/code/session_01JV7FJLEKixpyEzkRcru4Pj
…eneration - Audit now promotes recurring errors (3+ in CLAUDE_ERRORS.md) to practices/inbox - Audit now captures scoring gaps as practices for the pipeline to process - update-practices generates rules files when a practice warrants a repeatable constraint Closes: Aprendizaje→Contexto, Memoria→Aprendizaje, Auditoría→Aprendizaje https://claude.ai/code/session_01JV7FJLEKixpyEzkRcru4Pj
- New skills: watch-upstream and scout-repos (were stubs, now formal)
- New: practices/sources.yml with curated repos for scout
- Agent memory operationalized: 4 agents read/write .claude/agent-memory/
- Score trending: audit appends history entries to registry
- Fix: {{CLAUDE_KIT_PATH}} resolution documented in global sync
https://claude.ai/code/session_01JV7FJLEKixpyEzkRcru4Pj
- Add template/agent-memory/.gitkeep so bootstrap has the dir structure - Bootstrap Step 9b now seeds individual agent .md files - Update practices/README.md with all intake channels (watch, scout, audit gaps) - Expand sources.yml with 4 community repos for /forge scout https://claude.ai/code/session_01JV7FJLEKixpyEzkRcru4Pj
- Add MIT LICENSE and CONTRIBUTING.md
- Rewrite README.md with full architecture, stacks, skills, agents docs
- Replace all hardcoded ~/Documents/GitHub/claude-kit/ with $CLAUDE_KIT_DIR
- Sanitize registry/projects.yml (empty with example schema)
- Generalize practices (remove project-specific names)
- Translate CLAUDE_ERRORS.md to English
- Clean personal references from docs, agents, stacks, skills
- Fix {{CLAUDE_KIT_PATH}} → $CLAUDE_KIT_DIR in global settings template
- detect-claude-changes.sh now auto-resolves CLAUDE_KIT_DIR from script location
https://claude.ai/code/session_01JV7FJLEKixpyEzkRcru4Pj
All 13 doc files now have both English and Spanish sections with language switcher at the top. Pattern: EN first, divider, ES second. Files fully bilingual: - README.md, CONTRIBUTING.md, CLAUDE_ERRORS.md - docs/: best-practices, security-checklist, prompting-patterns, creating-stacks, troubleshooting, anatomy-claude-md, memory-strategy - practices/README.md Files with bilingual header note (technical logs): - docs/changelog.md, docs/roadmap.md, docs/analysis-v1.2.1.md https://claude.ai/code/session_01JV7FJLEKixpyEzkRcru4Pj
Add "Why claude-kit" sections to README (EN/ES) with competitive positioning table, comparison matrix, and 6 unique attributes. Add vision statement to roadmap. Update CONTRIBUTING intro with end-to-end positioning. https://claude.ai/code/session_01JV7FJLEKixpyEzkRcru4Pj
Roadmap informed by analysis of 10+ competitive repos (superpowers, trailofbits, claude-health, awesome-toolkit, developer-kit, etc.). Key themes per version: - v1.2.3: security hardening (prompt injection audit, hook profiles, TDD warning) - v1.3.0: stack expansion (node, java, aws, go) + cross-tool export - v1.4.0: plugin packaging for marketplace distribution - v1.5.0: session insights and analytics (markdown, not dashboards) https://claude.ai/code/session_01JV7FJLEKixpyEzkRcru4Pj
Colloquial, benefit-focused explanation of every roadmap item for quick reference and team onboarding. https://claude.ai/code/session_01JV7FJLEKixpyEzkRcru4Pj
Maps each roadmap item to affected files, dependencies, risk levels, and recommended implementation order. https://claude.ai/code/session_01JV7FJLEKixpyEzkRcru4Pj
- Add prompt injection detection as audit item 12 (scan rules/CLAUDE.md for suspicious patterns) - Add hook profiles (FORGE_HOOK_PROFILE: minimal/standard/strict) to block-destructive.sh - Add Type column (syntax|logic|integration|config|security) to CLAUDE_ERRORS.md format - Add git worktree isolation for Agent Teams in agents.md and implementer.md - Add TDD warning hook (warn-missing-test.sh) for strict profile - Update scoring formula for 12 recommended items (preserves 70/30 split) - Bump version to 1.2.3 https://claude.ai/code/session_01JV7FJLEKixpyEzkRcru4Pj
New stacks (5): - node-express: Node.js + Express/Fastify rules and permissions - java-spring: Java + Spring Boot + Maven/Gradle rules and permissions - aws-deploy: AWS CDK/SAM/CloudFormation rules with deny list for destructive ops - go-api: Go modules + standard library HTTP rules and permissions - devcontainer: Devcontainer configuration rules for Claude Code in containers New features: - /forge export skill: export config to Cursor (.cursorrules), Codex (AGENTS.md), Windsurf (.windsurfrules) - Bootstrap profiles: --profile minimal|standard|full controls what gets installed - Project tier detection in audit: simple/standard/complex adjusts scoring expectations - Audit item 12 (prompt injection) now reflected in audit-project skill Updated: - stacks/detect.md with 5 new detection rules - forge.md dispatcher with export command and bootstrap --profile - CLAUDE.md architecture section with new stacks and export command - Bump version to 1.3.0 https://claude.ai/code/session_01JV7FJLEKixpyEzkRcru4Pj
- Add .claude-plugin/plugin.json with full metadata for Claude Code plugin system - Add .claude-plugin/INSTALL.md documenting plugin vs full installation modes - Add plugin.json to all 13 stacks for standalone distribution (claude-kit-stack-python-fastapi, claude-kit-stack-react-vite-ts, etc.) - Stack plugins are composable: multiple can be installed, permissions merge via union - Plugin mode = curated subset (hooks + rules + commands) - Full mode = git clone + sync.sh (skills, agents, practices pipeline) - Bump version to 1.4.0 https://claude.ai/code/session_01JV7FJLEKixpyEzkRcru4Pj
- Add /forge insights skill: analyzes sessions for error patterns, file activity, agent usage, score trends. Generates recommendations and feeds practices pipeline. - Add session-report.sh Stop hook: generates SESSION_REPORT.md on session end (opt-in via FORGE_SESSION_REPORT=true) - Add scoring trends to /forge status: ASCII sparkline, trend arrows, alerts when score drops >1.5 points, sync recommendations for low scores - Update forge.md dispatcher with insights command - Update plugin.json with session-report hook and version bump - Bump version to 1.5.0 https://claude.ai/code/session_01JV7FJLEKixpyEzkRcru4Pj
- v1.6.0 (Ecosystem & Automation): MCP server templates, CI integration (GitHub Action + GitLab CI), stack auto-update detection - v1.7.0 (Team & Governance): team mode with config inheritance (base → team → individual), model routing rules per agent - Update roadmap state to v1.5.0 https://claude.ai/code/session_01JV7FJLEKixpyEzkRcru4Pj
- Changelog: add entries for v1.2.3, v1.3.0, v1.4.0, v1.5.0 - README: update both EN/ES sections — 13 stacks, 11 skills, 12 audit items, project tier, new commands (export, insights), version references - Security review: repo clean for public release (no secrets, no PII, no hardcoded paths, registry empty, .gitignore covers sensitive files) https://claude.ai/code/session_01JV7FJLEKixpyEzkRcru4Pj
14 test sections covering all features from v1.2.3–v1.5.0: hook profiles, TDD warning, bootstrap profiles, stack detection, audit tier + prompt injection, cross-tool export, plugin packaging, session insights, session report, scoring trends, error classification, git worktree refs. Estimated ~48 min total execution. https://claude.ai/code/session_01JV7FJLEKixpyEzkRcru4Pj
When running as root (sudo, containers, Claude Code web), $HOME points to /root instead of the actual user's home. The script now resolves the correct target via three methods in order: 1. SUDO_USER env var (set by sudo) 2. File owner of the claude-kit directory 3. Infer from repo path under /home/<user>/ CLAUDE_HOME env var still works as explicit override. https://claude.ai/code/session_01JV7FJLEKixpyEzkRcru4Pj
- Platform detection: Linux, macOS, WSL, Git Bash - Symlink fallback: auto-detects if symlinks work on Windows, falls back to file copies with source markers if not - Home directory resolution works on all platforms: - SUDO_USER for sudo on Linux/macOS - File owner detection (stat -c on Linux, stat -f on macOS) - Path inference for /home/<user>/ and /Users/<user>/ - CLAUDE_HOME env var as explicit override - Output shows platform and method (symlinks vs copies) - Messages in English for consistency - Updated INSTALL.md with per-platform instructions and troubleshooting table https://claude.ai/code/session_01JV7FJLEKixpyEzkRcru4Pj
Claude Code requires YAML frontmatter with name/description fields to recognize custom slash commands. Without it, /forge shows "Unknown skill". https://claude.ai/code/session_01JV7FJLEKixpyEzkRcru4Pj
Ensures /forge is available within the claude-kit repo itself, not just via global symlink. https://claude.ai/code/session_01JV7FJLEKixpyEzkRcru4Pj
Creates .claude/agent-memory/ with seeds for the 4 agents that have memory: project (implementer, architect, code-reviewer, security-auditor). Closes audit gap item 9 → score 10.0. https://claude.ai/code/session_01JV7FJLEKixpyEzkRcru4Pj
Creates 15 individual command files (forge-audit, forge-sync, forge-bootstrap, etc.) that appear in Claude Code's / autocomplete with descriptions. The original /forge <arg> still works too. Also updates sync.sh to install all commands from global/commands/ (not just forge.md), so new subcommands propagate automatically. https://claude.ai/code/session_01JV7FJLEKixpyEzkRcru4Pj
Remove forge.md and forge-*.md from .claude/commands/ (project scope). These belong exclusively in global/commands/ and are installed to ~/.claude/commands/ via sync.sh. Project .claude/commands/ now only has development commands: audit.md, debug.md, health.md, review.md Global global/commands/ has user-facing forge commands: forge.md + 15 forge-*.md subcommands https://claude.ai/code/session_01JV7FJLEKixpyEzkRcru4Pj
Claude Code does not follow symlinks for command discovery. sync.sh now copies command files to ~/.claude/commands/ and uses diff to detect updates. Skills and agents keep using symlinks. This ensures /forge-* commands are available in all projects, including claude-kit itself (dog-fooding). https://claude.ai/code/session_01JV7FJLEKixpyEzkRcru4Pj
First entry in projects.yml. Fixed legacy scoring formula in registry comment (was *0.5, correct is *(3.0/7)). https://claude.ai/code/session_01JV7FJLEKixpyEzkRcru4Pj
Covers three scenarios: new projects, existing projects without claude-kit, and projects already using claude-kit. Includes full command reference, stack detection, audit scoring, practices pipeline, bootstrap profiles, generated structure, and FAQ. https://claude.ai/code/session_01JV7FJLEKixpyEzkRcru4Pj
luiseiman
added a commit
that referenced
this pull request
Apr 21, 2026
v3 behavior `verify-before-done` trigger #1 regex enumerates major language test runners but omits `bash tests/*.sh` — dotforge's own convention. Reproduced during v3.3.0 push: legitimate verification via bash tests doesn't set the `verification_done` flag, soft-blocks every git push. Captured for next /forge update cycle. Proposed fix: add `bash\s+\S*test\S*\.sh` pattern to the regex. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
luiseiman
added a commit
that referenced
this pull request
Jun 4, 2026
…haviors v3 (#5) Step #1 of the native-first plan completed. Portfolio scan: 0 overrides across all 12 projects over ~7 weeks (behaviors live since v3.0), including production. Only 4/12 adopted behaviors. Verdict in native-vs-dotforge-boundary.md is now SPLIT: - KEEP the graduated escalation engine (soft_block) — exercised and works; the happy path is "verify" not "override" (observed live this session). - RETIRE/SIMPLIFY the auditable override trail + v4 capture loop — the v3-vs-hookify differential is empirically refuted (0 uses); the loop processes an always-empty log. Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
luiseiman
added a commit
that referenced
this pull request
Jun 4, 2026
…tion (#6) Step #1 validation found 0 overrides across all 12 projects in ~7 weeks (production included; only 4/12 adopted behaviors). The auditable override trail — the v3-vs-hookify differential — has no empirical use, and the v4 capture loop processed an always-empty log. Retired as dead weight. The graduated escalation engine stays: it is exercised and the happy path is "verify", not "override". Removed: - scripts/process-override-log.sh, tests/test-process-override-log.sh - template/hooks + .claude/hooks session-start-process-overrides.sh - .forge/audit/overrides.log (git-tracked) - lib.sh: forge_override_append, FORGE_AUDIT_LOG, audit-log append in try_override - SessionStart wiring in settings.json + template/settings.json.tmpl - audit Dimension B item B3 (override loop); B now 4 items (forge_adoption 0-4) Kept: - forge_pending_block_try_override (override-via-reinvocation, state.json only) - full soft_block escalation Updated both scoring engines (score.sh + audit_all.py), checklist/scoring/SKILL, registry, CLAUDE.md, README, usage-guide, guia-uso. 19 tests green. Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.