v0.14.1

Features

• statusline rewrite — richer StatusLine output with new segments and accurate context tracking:
  • 1M-context awareness — detects Opus 4.7 [1m] extended mode and scales the context window to 1,000,000 tokens (falls back to 200k otherwise)
  • context usage bar — 8-segment colored bar (🟢 🟡 🟠 🔴 🚨) driven by the latest assistant usage from the transcript, with AUTO-COMPACT! / LOW! alerts surfaced from system_message entries
  • session metrics — cost (💰, colored by threshold and formatted in ¢ under $0.01), duration (⏱), and net lines (📝 ±N) pulled from the hook JSON
  • relative directory — shows the current dir relative to project root when nested
  • richer git segment — branch with change count, colored red when dirty / green when clean
  • graceful jq fallback — prints an install hint and exits cleanly when jq is missing instead of producing a broken line

Docs

• /setup-context — StatusLine prompt updated to describe the new segments and notes the jq runtime dependency
• README — StatusLine description updated to match the new output

Full Changelog: v0.14.0...v0.14.1

v0.14.0

0.14.0 (2026-03-06)

Features

• project-local hooks — notification and stop hooks now configured in .claude/settings.json (project) instead of ~/.claude/settings.json (global)
• tool failure guard — new PostToolUseFailure hook that stops Claude after 4+ consecutive failures, triggers error sound + notification + AskUserQuestion
• batch-operations skill — 4-step workflow (scope, batch, verify, clean up) for large refactors
• git-platform detection skill — auto-detects GitHub/GitLab/Azure DevOps/Bitbucket from remote URL
• multi-platform commands — create-pr, pr-comment, code-review, fix-bug now detect git platform
• improved fix-bug investigator — stack trace analysis, binary search debugging, type tracing, intermittent bug handling
• .claudeignore template — shipped on init with sensible defaults for all project types
• CLAUDE.md efficiency rules — read before changing, follow existing patterns, code only, progressive loading, targeted tests
• CLAUDE.md compact instructions — what to preserve/remove during context compaction

Changes

• doctor now checks project-local hooks and warns about legacy global hooks
• notify.sh and tool-failure-guard.sh installed to project .claude/scripts/

v0.13.3

0.13.3 (2026-03-05)

Changes

• context - update CONTEXT.md with modular src/ directory structure
• decisions - add ADR-012 (Agent File Extraction) and ADR-013 (Structured Exploration)
• mcp - add .mcp.json with context7 MCP server configuration

v0.13.2

0.13.2 (2026-03-05)

Fixes

• zsh completion — replaced _dotcontext "$@" with compdef _dotcontext dotcontext to prevent _arguments error when eval'd
• shell auto-detection — dotcontext completion without arguments now detects zsh vs bash automatically via $ZSH_VERSION

v0.13.1

0.13.1 (2026-03-05)

Docs

• README — added completion to CLI commands table, /release to Claude Code commands table, new "Shell Completion" section with bash/zsh setup instructions

v0.13.0

0.13.0 (2026-03-05)

Features

• modular source architecture — split monolithic script into 14 src/ modules (core/, commands/, setup/) with Makefile build system; still ships as single executable
• colored help screen — grouped layout with BLUE BOLD headers, CYAN commands, YELLOW options, and adaptive column widths based on terminal width
• icon system — 16 ICON_* constants replacing hardcoded symbols across doctor, init, and update output; structured output blocks with print_block_header/print_block_footer
• shell tab completion — dotcontext completion bash|zsh generates working completions with per-subcommand option awareness
• TTY-safe colors — [[ -t 1 ]] detection strips escape codes when output is piped

Changes

• ADR-001 v2.0 — updated to reflect modular source with build-time bundling
• CLAUDE.md — architecture section updated, completion command documented

V0.12.0

Features

• /deep-context restructured to 4-step exploration — Replaced the 5-agent model with a structured 4-step progression:
  • Step 1: Overview Agent — architecture summary, key files, entry points
  • Step 2: Subsystem Agent — module map, interdependencies, boundaries
  • Step 3: Drill Agent — targeted deep-dive into relevant areas
  • Step 4: Data Flow Agent — trace information movement through the system
  • Steps 1+2 run in parallel, Steps 3+4 run sequentially
• Declarative managed directory cleanup — New cleanup_managed_dir() removes stale files from managed-only directories using a desired-state allow-list, eliminating future migration code
  • Guards against empty expected list and preserves symlinks
  • Excludes .claude/commands/ (users create custom commands there)
• ADR-013: Structured Exploration Pattern

Fixes

• Stale agent references — dotcontext init and update referenced deleted agent files; updated to new step-based agents
• Agent count — Init output corrected from 13 to 12

Changes

• README — Skills directory corrected, MCP server config documented, deep-context architecture updated

Full Changelog: v0.11.0...v0.12.0

v0.11.0

Features

• /commit command — Smart commit workflow with style-aware message generation:
  • detects project commit style from git log (conventional commits vs freeform)
  • interactive staging via AskUserQuestion (multiSelect, directory grouping)
  • diff analysis and AI-generated commit messages
  • user confirmation before committing, --amend support
• dotcontext doctor — New CLI subcommand for project health validation:
  • 11 checks: Claude CLI, .context/ structure, CLAUDE.md, CONTEXT.md, decisions, commands, agents, skills, MCP config, notification hooks, git status
  • colored pass/fail/warn output with summary line
• StatusLine — At-a-glance project context in Claude Code:
  • script showing model, context %, git branch+changes, .context health
  • configured via new task 8 in /setup-context
  • downloaded during dotcontext init, managed on update
• Agent extraction — 13 inline agent prompts extracted to reusable .claude/agents/ files:
  • /code-review: compliance-checker, bug-detector, security-analyst (3 agents)
  • /deep-context: scope-guardian, primary-explorer, cross-repo-explorer, cross-repo-validator, reviewer-output (5 agents)
  • /fix-bug: investigator, fix-conservative, fix-minimal, fix-refactor, reviewer (5 agents)
  • commands now reference agents via Read .claude/agents/... — behavior unchanged
  • agents included in managed templates for init and update
• ADR-012: Agent File Extraction Pattern — Documents the file-based agent architecture
• Update safety — exec re-exec after CLI update ensures templates use the new managed list in a single dotcontext update run

Changes

• Skills directory migration — Moved .context/skills/ to .claude/skills/ with updated references across commands and templates

v0.10.2

Docs

• Rebrand from "scaffold" to "toolkit" across all user-facing text (README, CLAUDE.md, install screen, CLI help, CONTEXT.md)

Full Changelog: v0.10.1...v0.10.2

v0.10.1

Docs

• Collapse quickstart gif in README behind <details> for a cleaner first impression

Full Changelog: v0.10.0...v0.10.1