Changelog

All notable changes to haven are documented here.

Format: Keep a Changelog. Versioning: Semantic Versioning.

[v0.9.0] — 2026-03-27

Added

  • haven telemetry subcommand — telemetry management is now a proper subcommand (haven telemetry status, haven telemetry on/off). Top-level shortcuts haven note, haven bug, and haven question are also added for quick feedback capture.

  • haven status --verbose — shows internal state detail (lock files, paths, raw state) useful for debugging.

  • --filter and --count flags on haven list — filter listed files by pattern and/or print only a count.

  • haven update <file> — top-level alias for haven add --update.

  • --cmd-name flag on haven completions — allows generating completions for an alias (e.g. h) instead of the default binary name.

Fixed

  • haven ai remove accepts source URL — haven ai remove now matches by source URL (or suffix) in addition to skill name, and shows available skill names when no match is found.

  • Local platforms registry path — haven now reads the local platforms registry from ~/.haven instead of the old ~/.dfiles path.

  • Atomic writes for state files — state.json and haven.lock are now written via temp+rename to prevent partial writes on crash.

  • apply.lock race condition — lock file creation now uses O_EXCL (create-new) to eliminate a TOCTOU race.

  • Process liveness check — replaced the kill(0) subprocess approach with the nix crate for cleaner cross-platform signal handling.

  • Dry-run output memory leak — removed .leak() call in dry-run output path.

  • Stale dfiles references — remaining internal comments and test names referencing dfiles have been updated to haven.

[v0.8.0] — 2026-03-27

Added

  • CLAUDE.md staleness detection — haven status now shows a ~ marker in the [ai] section when the haven-managed section in ~/.claude/CLAUDE.md is out of date. This fires when a skill snippet file (all.md / claude-code.md) has been edited since the last haven apply, or when a new skill has been added but not yet deployed. Run haven apply --ai to bring CLAUDE.md up to date.

  • Snippets at the top of the managed section — skill snippets from ai/skills/<name>/all.md and ai/skills/<name>/claude-code.md now appear before the skills/commands listing inside the haven-managed block. Claude Code reads snippets earlier in the file, ensuring they take effect before the index.

Fixed

  • False drift on CLAUDE.md — haven status and haven diff no longer report M or MC drift on ~/.claude/CLAUDE.md caused by the haven-generated section that haven apply --ai appends. Only user content outside the haven markers is compared. Previously, every haven status after an --ai apply would show the file as modified.

  • Legacy <!-- dfiles managed --> markers removed — the old pre-rename marker strings (<!-- dfiles managed start/end -->) that appeared in platform config injection have been replaced with <!-- haven managed start/end --> throughout. If your CLAUDE.md contains the old dfiles markers, run haven apply --ai to migrate to the new format. (CLAUDE.md is now fully managed by haven apply --ai via a single unified section; the old dual-section approach is gone.)

[v0.7.6] — 2026-03-26

Added

  • runner accepts an array — the runner field in ai/config.toml now accepts either a string or an array, letting you invoke agent-skills-cli via a package runner without a global install: toml [skills] backend = "agent-skills" runner = ["bunx", "agent-skills-cli"] # or ["npx", "agent-skills-cli"] runner = "skills" # string shorthand still works

  • haven upgrade sudo fallback — when the upgrade binary write fails with a permission error (e.g. haven is installed in /usr/local/bin), the command now detects this, prints a clear message, and asks: error: Permission denied writing to /usr/local/bin/haven. Retry with sudo? [y/N] If you confirm, it runs sudo mv + sudo chmod 755 to complete the install. The download and checksum steps are not repeated. The extracted binary is staged in /tmp rather than a sibling .new file, so the permission error is caught at the move step rather than the extract step.

Fixed

  • haven ai backends no longer silently shows native — previously, any error loading ai/config.toml (e.g. a parse error or missing file) would silently fall back to the native backend, making it appear as the active backend even when another was configured. The error is now propagated so you see what went wrong.

[v0.7.5] — 2026-03-26

Added

  • AgentSkills backend — new backend = "agent-skills" in ai/config.toml delegates skill fetch and deployment to agent-skills-cli. Gives access to 175K+ marketplace skills with cross-agent deployment (-a flag). Requires npm install -g agent-skills-cli (Node.js 18+). Haven retains full control of CLAUDE.md generation, state.json ownership, and collision detection. Configuration: toml [skills] backend = "agent-skills" runner = "skills" # default; accepts full path to binary timeout_secs = 120 # default Search (haven ai search) routes to the agent-skills marketplace when this backend is active.

Removed

  • SkillKit backend — the skillkit backend has been removed. The native backend remains the only supported backend. Any ai/config.toml with backend = "skillkit" will error with an "unknown skill backend" message; remove that line or set backend = "native" to continue.

[v0.7.1] — 2026-03-26

Fixed

  • hv upgrade checksum verification — the upgrade command was requesting SHA256SUMS but the release workflow publishes haven-vX.Y.Z-SHA256SUMS. This caused all self-upgrade attempts from v0.7.0 to fail with a 404 error.

[v0.7.0] — 2026-03-26

Added

  • Conflict detection — haven apply --files now detects when you've edited a deployed file since the last apply and asks what to do: [s]kip, [o]verwrite, [A]pply all, or [d]iff (view a diff before deciding). Haven records a SHA-256 fingerprint of every file it writes, so it can tell the difference between "I updated this in source" and "you edited this live copy". The --on-conflict=<mode> flag skips the prompt: skip (CI-friendly, exits 1 when anything was skipped), overwrite (always clobber), or prompt (default on a TTY).

  • C marker in haven status — files you've edited since the last apply now show a C marker. Combined with the source-drift marker: MC means both the source and your live copy have diverged. Run haven status before haven apply to see exactly what you've locally modified.

[v0.6.0] — 2026-03-24

Added

  • Swappable skill backends — the AI skills pipeline now delegates to a configurable backend. Configure via ai/config.toml: toml [skills] backend = "native" # default Existing repos need no changes — the native backend is the default.

  • haven ai backends — new subcommand that lists all known backends with availability status and the currently active backend marked.

  • docs/reference/skill-backends.md — new reference page covering the native backend configuration.

[v0.5.5] — 2026-03-24

Added

  • Brewfile auto-sort — set [homebrew] sort = true in any module config to keep the Brewfile sorted alphabetically after every haven brew install or haven brew uninstall. Each kind (tap, brew, cask) is sorted independently; blank lines and comments are preserved in place.
  • haven apply Brewfile summary — the apply summary now reports how many Brewfiles were run: Applied N file(s), M Brewfile(s) across K module(s).
  • haven status improvements — the profile name is now printed at the top of status output. A ✓ Everything up to date message is shown when no drift is found.
  • haven import symlink warning — after importing from chezmoi, a summary of all imported symlink_ entries is printed with a reminder to verify symlink targets before running haven apply.
  • Documentation site — full MkDocs documentation at johnstegeman.github.io/haven, covering concepts, task-oriented guides, a reference section, and a dedicated "For chezmoi Users" section. Deployed automatically via GitHub Actions.

Fixed

  • Symlink idempotency — haven apply no longer re-writes a symlink that already points to the correct target. Previously it would overwrite and count the file even when nothing changed.
  • Dangling symlink backup — haven apply no longer attempts to back up a dangling symlink (one whose target has been deleted) before overwriting it. std::fs::copy follows symlinks and would fail; the dangling link is now removed directly.
  • brew bundle install output — spurious Using <formula> lines from brew bundle are now suppressed. Output is streamed in real time instead of being buffered until completion.
  • Unused import — removed unused serde_yaml import in chezmoi.rs.

[v0.5.0] — 2026-03-23

Added

  • haven apply --zap — when removing unreferenced casks, also delete their associated app data and support files (brew uninstall --cask --zap). Implies --remove-unreferenced-brews.
  • haven telemetry --action/--bug/--question "<text>" — typed telemetry annotations with auto-generated, sequenced IDs (A000001, B000001, Q000001). The ID is printed after the command so you can reference it in follow-up notes. Complements --note (which uses N prefix).
  • haven telemetry --list-notes/--list-bugs/--list-actions/--list-questions — filter the telemetry log to a specific annotation kind.
  • haven list — rewritten to show files, Homebrew packages, and AI skills in a unified view. [files], [brew], and [ai] section headers appear when no filter is active. --files, --brews, and --ai flags scope the output to one section. --profile <p> scopes to a specific profile.
  • haven status --brews — now shows both missing packages (?) and packages installed but not in any Brewfile (+), giving a unified install/extra drift view in one pass.

Fixed

  • haven apply file counter (B000002) — the Applied N file(s) summary now counts only files actually written. Previously it counted every entry processed, including files silently skipped because the destination was already identical to the source.
  • haven apply --brews — now runs brew bundle install for the master Brewfile and every active module Brewfile. Previously only the master brew/Brewfile was used, silently ignoring module-level packages.
  • brew leaves tap-qualified names — haven status --brews and haven apply --remove-unreferenced-brews no longer report tap formulae (e.g. qmk/qmk/qmk) as extra when their short name (qmk) is declared in a Brewfile. Matching is now done by short name in both directions.
  • haven brew install — when there is no master brew/Brewfile but exactly one module Brewfile exists, it is used automatically. If multiple module Brewfiles exist and --module is not given, haven errors with a clear hint.
  • haven init — scaffold always appends a [profile.default] section to haven.toml if one is not present, so bare haven apply works out of the box.
  • Removed --no-lock from brew bundle install — the flag was removed from modern Homebrew and caused an error on recent versions.

[v0.4.0] — 2026-03-23

Breaking

  • Project renamed from dfiles to haven — the binary, config file, state directory, and all environment variables have changed:
Before After
dfiles (binary) haven
dfiles.toml haven.toml
~/.dfiles/ ~/.haven/
DFILES_DIR HAVEN_DIR
DFILES_VCS HAVEN_VCS
DFILES_TELEMETRY HAVEN_TELEMETRY
DFILES_PROFILE HAVEN_PROFILE
GitHub repo johnstegeman/dfiles johnstegeman/haven

The source encoding (dot_, private_, .tmpl, etc.) and repo layout are unchanged — existing repos continue to work by moving dfiles.toml → haven.toml and ~/.dfiles/ → ~/.haven/.

Added

  • config/ignore is now a Tera template — the ignore file is rendered against the current machine context (os, hostname, username, profile, data.*) before patterns are evaluated. This matches how chezmoi treats .chezmoiignore: conditional patterns like {% if os == "macos" %}.DS_Store{% endif %} work as expected. If rendering fails, haven warns and falls back to ignoring nothing. (haven import --from chezmoi now converts Go template syntax in .chezmoiignore to Tera syntax rather than stripping template lines.)

Fixed

  • haven init in an already-initialised repo no longer errors — it prints an informational message and exits 0. The VCS resolution prompt (Jujutsu vs Git) now only runs when a source URL is provided, not during a blank scaffold init.
  • haven --version now shows the full <semver>+<short-commit> build identity (e.g. 0.4.0+abc1234), matching the version field written to telemetry events.

[v0.3.0] — 2026-03-23

Security

  • Fixed HIGH: tarball symlink traversal in extract_tarball — a malicious gh: skill package could plant a symlink entry pointing outside the cache directory, then write a file through it on extraction. Symlink and hard-link tar entries are now unconditionally skipped before extraction. Users on v0.2.0 who fetch skills from untrusted gh: sources should upgrade before running haven apply. See .gstack/security-reports/2026-03-23.json for the full audit record.

Added

  • haven security-scan — audit all tracked source files for secrets, sensitive filenames, and credential paths. Checks filename patterns (.env, id_rsa, .pem…), sensitive paths (~/.aws/credentials, ~/.kube/**, ~/.ssh/**…), and content patterns (GitHub tokens, AWS keys, PEM private keys, OpenAI/Anthropic API keys, generic password assignments). Optional --entropy flag adds high-entropy string detection. Exits 1 when findings are found (CI-friendly). False positives suppressed via [security] allow in haven.toml.
  • haven add content scan — when adding a file, its content is automatically scanned for secrets. A prompt is shown before saving; declining removes the file from source/ with no partial state left behind.
  • haven completions <shell> — print shell completion scripts to stdout. Supports fish, zsh, and bash. All subcommands and flags are included.
  • haven list — list all tracked files with decoded destination paths and flag annotations (template, private, symlink, extdir, extfile, etc.).
  • haven telemetry — manage local telemetry from the CLI. --enable / --disable flip [telemetry] enabled in haven.toml using surgical toml_edit (preserves comments and formatting). --note "<text>" appends a {"kind":"note"} entry to ~/.haven/telemetry.jsonl regardless of whether telemetry is enabled — useful for annotating test runs, onboarding sessions, or observed issues so the log has context during analysis. Bare haven telemetry prints the current enabled/disabled status.
  • haven upgrade — self-update command. Downloads the latest release tarball from GitHub, verifies the SHA256 checksum, extracts the binary, and atomically replaces the running executable. --check flag exits 0 when up to date, 1 when an update is available (CI-friendly). Supports macOS (arm64, x86_64) and Linux (x86_64, aarch64, armv7, i686 musl).
  • Jujutsu (jj) VCS backend — configure haven to use jj git clone --colocate for all new clone and init operations. Set via --vcs jj flag, HAVEN_VCS=jj env var, or [vcs] backend = "jj" in haven.toml. On first use without a config, haven detects whether jj is on your PATH and prompts once. haven vcs shows the active backend and how it was resolved.
  • haven unmanaged — walk ~ and report files not tracked by haven. Only dotfiles and dotdirs are examined at the home root. High-noise directories (.cache, .cargo, node_modules, .git, Library, etc.) are skipped. --path <dir> scans a specific directory; --depth <n> controls recursion depth (default: 3). Useful for discovering dotfiles to add.
  • [data] custom template variables — define arbitrary string variables in haven.toml under [data] (e.g. host = "my-laptop"). Available in all .tmpl files as {{ data.host }}. haven import --from chezmoi automatically migrates .chezmoidata.yaml / .chezmoidata.toml into [data] entries. haven data prints all resolved variables.
  • haven data — show all template variables in scope: built-in variables (os, hostname, username, home_dir, source_dir) and custom [data] entries from haven.toml. Useful for debugging templates.
  • config/ignore warning on haven add — if a file being added matches an ignore pattern, the add is skipped with a clear message explaining how to remove the pattern.

Added (docs)

  • docs/from-chezmoi.md — getting started guide for chezmoi users. Covers the automated migration path, template syntax conversion table, command equivalence table, Brewfile/module/profile setup, AI skill import, new-machine bootstrap, daily workflow comparison, and a gap table of chezmoi features not yet in haven.

Fixed

  • Installer URL typo (affected the guide; the binary download URL was correct).
  • Pre-existing build warnings: unused import and dead code in internal modules.

[v0.2.0] — 2026-03-21

Added

  • haven ai add-local — import a locally developed skill directory into the haven repo. Moves the skill into ai/skills/<name>/files/, writes skill.toml with source = "repo:", creates a blank snippet stub, and removes the original directory. Run haven apply --ai afterward to deploy.
  • extfile_ source encoding — track single-file and archive remote downloads in source/. Two types: type = "file" (plain download) and type = "archive" (tarball extract with optional subpath). SHA-256 verification supported. haven diff shows ? when the destination is absent.
  • haven ai search — search the skills.sh registry for available skills. Results show source in gh:owner/repo/skill format and install count.
  • haven ai scan — scan an existing skills directory for unmanaged skills and offer to add them to ai/skills.toml. Detects source via git remote or skills.sh fuzzy search.
  • Managed config section injection — haven apply --ai injects skill snippets from all.md / <platform>.md into platform config files (e.g. ~/.claude/CLAUDE.md) between HTML comment markers. Idempotent.
  • haven diff --ai — shows stale skill SHA drift and missing deployments.
  • haven diff extdir ref drift — shows when a cloned external repo is behind the pinned ref in the marker file.
  • haven.lock SHA verification — on cache miss, the freshly-fetched SHA is compared against the lock entry. Mismatch is a hard error; run haven ai update to accept intentional upgrades.
  • Parallel skill fetches — haven apply --ai fetches multiple gh: skills concurrently using std::thread::scope. A single miss fetches inline; 2+ prints "Fetching N skills in parallel…" with per-skill results.
  • Apply file lock — writes PID to ~/.haven/apply.lock to prevent concurrent runs. Stale locks (dead PID) are cleaned up automatically.
  • haven source-path — print the repo directory path. Useful in scripts and shell aliases.
  • XDG default repo directory — new installs default to ~/.local/share/haven. Existing ~/haven repos are still detected and used without migration.
  • haven add --update — re-copy a file into source/ even if already tracked.
  • home_dir template variable — available in .tmpl files alongside hostname, username, os, etc.
  • create_ prefix support — create_ files are written on apply only if the destination does not already exist.
  • exact_ directory prefix — on apply, files in the destination that are not tracked in source/ are removed (backed up first).
  • Script execution — haven apply --run-scripts runs scripts from source/scripts/. run_once_ / once_ scripts are tracked in state and skipped on subsequent runs.
  • Opt-in local telemetry — disabled by default. When enabled, events are written to ~/.haven/telemetry.jsonl (command name, flags, duration, exit status — no paths or personal data).

Fixed

  • modify_ scripts in chezmoi import now emit a clear skip message with guidance instead of silently disappearing.
  • Config structs use forward-compatible deserialization (unknown fields are ignored rather than erroring).

[v0.1.0] — 2026-03-20 (initial release)

Added

  • haven init — create a blank scaffold or clone an existing haven repo. Supports gh:owner/repo[@ref] shorthand, --apply, --profile, --branch.
  • haven add — track a dotfile by copying it into source/. Sensitive filename detection prompts before saving. Directory add: recursively adds files or tracks the directory as an extdir_ external git clone.
  • haven apply — copy source files to their destinations, install Homebrew packages via brew bundle, run mise, deploy AI skills. --dry-run mode, --files/--brews/--ai section filters, --remove-unreferenced-brews, --interactive, --apply-externals, --run-scripts.
  • haven diff — show file-level diff between source/ and live files. --stat, --color, --profile, --module flags. Exit 1 on drift.
  • haven status — concise drift summary with ✓ M ? ! markers.
  • haven brew install/uninstall — run brew operations and keep Brewfiles in sync in one step.
  • haven import --from chezmoi — migrate from chezmoi. Handles dot_, private_, executable_, symlink_ prefixes; converts Go templates to Tera; imports .chezmoiexternal.toml git entries as extdir_ markers; imports .chezmoiignore as config/ignore.
  • haven ai add/fetch/update/remove/discover — manage AI agent skills across platforms. Skills declared in ai/skills/<name>/skill.toml, fetched from gh:owner/repo[@ref], pinned by SHA in haven.lock.
  • Magic-name file encoding — dot_, private_, executable_, symlink_, extdir_, .tmpl suffix. Compatible with chezmoi's encoding.
  • Tera templates — .tmpl files rendered with hostname, username, os, arch, env, profile, source_dir variables.
  • 1Password integration — {{ op(path="op://vault/item/field") }} template function reads secrets at apply time.
  • Profiles — [profile.<name>] in haven.toml with modules list and optional extends inheritance.
  • Modules — group Homebrew packages and mise configs. --module flag on apply/diff to scope package operations.
  • extdir_ externals — track remote git repos as clone markers; cloned on haven apply.
  • Cross-platform binaries — macOS (x86_64, arm64) and Linux (x86_64, aarch64, i686, armv7 musl).
  • Shell installer — curl -fsSL .../install.sh | sh with SHA256 verification and automatic PATH detection.