Skip to content

CHANGELOG.md

Latest commit

 

History

History
398 lines (333 loc) · 29.6 KB

CHANGELOG.md

File metadata and controls

398 lines (333 loc) · 29.6 KB
Raw

Changelog

All notable changes to this project will be documented in this file.

[0.3.4] - 2026-06-21

Added

  • AI-agent commands are now hidden from history recall by default — Up-arrow recall and Ctrl+R reverse search show your own (human-typed) commands only, so agent-issued commands no longer clutter your prompt history. Toggle agent commands on/off live with Ctrl+A (Up-arrow recall) or Alt+A (Ctrl+R search), per-invocation with --include-agents, or permanently via search.recall_show_agents in config.toml.
  • suv backup [--out PATH] — write a snapshot of your history database to a chosen path (defaults to a timestamped file in the backups directory). A snapshot is also taken automatically before any destructive suv delete; pass --no-backup to skip it.
  • suv bookmark pick — interactive picker that recalls a saved bookmark directly into your shell prompt for editing.
  • "Failed only" search filter — press Ctrl+E in the search TUI or run suv search --failed to restrict results to commands that exited non-zero.
  • suv stats --human — human-only analytics that exclude agent activity; also available as the h toggle inside the stats TUI.
  • suv agent report --fail-on <low|medium|high|critical> — exits non-zero when findings meet or exceed the given risk level, for use in local CI checks and git hooks.
  • Configurable agent.prompt_capture_max_chars — control how much prompt text is captured per agent command. Default raised from 500 to 4000 characters.
  • Configurable redaction.extra_patterns — supply your own secret-matching regexes to redact in addition to the built-in patterns.
  • Idempotent JSONL import — suv import now deduplicates, so re-importing the same export no longer creates duplicate entries. Pass --allow-duplicates to opt out.
  • New risk patterns — git clean -f, world-writable chmod, and recursive chown -R are now flagged by risk assessment.

Changed

  • Search ranking overhauled — results must now contain your typed words as literal substrings; the previous loose character-subsequence matching is gone (so gco no longer matches git checkout). Results are ranked prefix matches first, then contiguous matches, then in-order words, then any-order words.
  • Up-arrow recall is now recency-first — a command typed in another directory is no longer buried after a cd; the most recently used commands surface first.
  • suv alias suggest is now human-only — alias suggestions are derived from your own commands and ignore agent activity.
  • alias add and add-suggested regenerate the alias file automatically — no separate regenerate step is needed after adding an alias.
  • MCP project-scoped tools now match the full directory subtree — queries scoped to a project include commands run in nested subdirectories.
  • MCP now honors mcp.default_days and mcp.default_limit — configured defaults are applied to MCP queries.
  • agent.risk_ignore_patterns now actually suppress matching risk findings — previously configured ignore patterns had no effect.
  • Dependencies — rusqlite 0.38 → 0.40, clap_mangen 0.2 → 0.3; lockfile refreshed.

Fixed

  • Redaction no longer corrupts legitimate command flags — docker -p, ssh -p, and curl -u are left intact instead of being mangled by secret redaction.
  • Password redaction edge cases — passwords containing @, and password-only database connection URIs, are now redacted correctly instead of leaking; base64 values following a key are now detected.
  • MCP multibyte panics — fixed UTF-8 byte-slice panics when agent output or prompts contained multibyte characters.

Security

  • Destructive prompts refuse non-TTY stdin — suv delete and suv uninstall no longer accept a piped y, preventing accidental destructive actions when input is not an interactive terminal.
  • SECURITY.md now documents the at-rest data model and redaction coverage — clarifying what is stored locally and what redaction does and does not cover.

[0.3.3] - 2026-04-25

Fixed

  • Cross-machine suv export → suv import no longer fails with FOREIGN KEY constraint failed — previously, a JSONL export only carried entries, so importing into a fresh database on another machine died on the very first row because the entry's session_id (and any tag_id) referenced rows that did not exist locally. The importer now auto-creates a placeholder session for any unknown session_id, and re-maps each entry's tag_id by looking up the carried-over tag_name on the destination — creating the tag locally when missing, falling back to a NULL tag association if the 20-tag cap is reached. Addresses #19.
  • suv import now rejects non-JSONL files up front with a clear message — passing a CSV (or JSON-array) export to suv import previously emitted a confusing per-line "expected struct Entry" parse error for every line. The importer now peeks the first non-empty line and aborts with a single actionable message pointing the user at suv export > history.jsonl. Addresses #19.
  • suv export --format json with zero matching entries now emits [] instead of an empty file — previously an empty result produced a zero-byte file that downstream JSON parsers rejected as invalid.

[0.3.2] - 2026-04-16

Added

  • Paste support in search TUI — Cmd+V / Ctrl+V now works in the search pane and all dialog inputs (filter, note, go-to-page). Pasted text is sanitized (control characters stripped), respects the 2000-character input limit, and routes to the active input field. In vim Normal mode, paste auto-switches to Insert mode. Closes #18.

Fixed

  • Reduced false positives in secret redaction — Environment variables like AUTHOR_NAME, GIT_AUTHOR_EMAIL, TOKENIZERS_PARALLELISM, PASSWORD_FILE, CREDENTIAL_HELPER, SECRET_SCANNING, and REACT_APP_AUTH_DOMAIN are no longer incorrectly redacted. The redaction engine now requires sensitive keywords (SECRET, TOKEN, PASSWORD, AUTH, etc.) to appear as the final segment of the variable name rather than matching as arbitrary substrings. Real secrets like GITHUB_TOKEN=, DB_PASSWORD=, API_KEY= are still correctly redacted. Closes #16.
  • Bash octal parsing error — Fixed value too great for base crash in bash hooks when EPOCHREALTIME milliseconds had a leading zero (e.g., 068). Added 10# prefix to force base-10 evaluation. #17.

[0.3.1] - 2026-04-05

Added

  • suv history — Non-interactive history dump with all standard filters (--after, --before, --tag, --exit-code, --executor, --here, --cwd). Supports -n for result count and --json for JSONL output. Pipeable to other tools. Newest-first by default.
  • suv doctor — Diagnostic command that checks shell version, shell hooks, config validity, database health (schema version + integrity check), recording state, MCP server registration (Claude Code and Cursor), and agent hook scripts. Reports pass/warn/fail with actionable fix hints.
  • Configurable search scoring — Three new [search] config fields: length_threshold (command length penalty, default 80), human_boost_percent (boost for human commands over agent commands, default 33%), cwd_boost_percent (boost for same-directory commands, default 50%). Tune via suv settings or config.toml. Setting a boost to 0 disables it.
  • pi.dev agent integration — suv init pi installs a TypeScript extension for pi.dev that captures bash commands and prompts via pi.dev's event system. Commands are recorded with executor=pi.

Changed

  • Full session UUID — Session filenames now use the full UUID instead of a truncated 8-character prefix, avoiding rare collisions.

[0.3.0] - 2026-04-01

Added

  • Agent Session Discovery — find_agent_session MCP tool searches past AI agent sessions by prompt text, directory, executor, or date range. Returns session summaries with command counts, success rates, risk breakdown, and claude --resume commands.

  • Session Replay — replay_agent_session MCP tool returns the full chronological timeline of a specific agent session with prompts interleaved between commands. Supports session ID prefix normalization (passing abc123 finds claude-abc123).

  • Learn from Failures — learn_from_failures MCP tool analyzes recurring command failures in a project. Shows commands with high failure rates, agent vs human failure comparison, and last failure timestamps. Helps agents avoid repeating known-bad approaches.

  • Project Context — project_context MCP tool returns a project briefing: common commands, build/test/lint patterns with success rates, recent failures, and agent activity. Available on-demand with directory and time range filters.

  • suvadu://agents/sessions resource — auto-injected summary of the 5 most recent AI agent sessions with prompts and command counts. Agents get session awareness before the first tool call.

  • suvadu://context/project resource — auto-injected project briefing at session start. Includes common commands, failure rates for frequent commands, and recent agent activity. Every new agent session starts informed.

  • MCP Configuration — new [mcp] section in config.toml and a new MCP tab in suv settings TUI. Disable individual tools or resources with checkboxes, set default time windows (default_days) and result limits (default_limit), and exclude directories from MCP queries. Config is loaded at MCP server startup; disabled tools/resources are hidden from agents.

Changed

  • MCP server expanded — 15 tools (was 11) and 7 auto-injected resources (was 5). New tools: find_agent_session, replay_agent_session, learn_from_failures, project_context. New resources: suvadu://agents/sessions, suvadu://context/project.

[0.2.1] - 2026-03-30

Added

  • Vim keybindings — optional vim-style modal navigation in the search TUI. Enable with vim_mode = true in [search] config or via suv settings → Search → Vim Mode. Insert mode for typing, Normal mode for j/k navigation, Ctrl+U/Ctrl+D half-page scroll, g/G jump to top/bottom, h/l page navigation, / or i to return to search, q to quit. Off by default. See #13.

Changed

  • Updated README — slimmed down from 840 lines to ~120 lines, linking to suvadu.sh for full documentation. Updated logo assets to new chevron stack design.
  • Updated website URL — all references now point to suvadu.sh instead of appachi.tech/suvadu.

[0.2.0] - 2026-03-25

Added

  • MCP Server — suv mcp-serve exposes shell history to AI agents via the Model Context Protocol. 11 tools including assess_risk for pre-execution safety checks. 5 browsable resources (history/recent, failures/recent, stats/today, risk/summary, agents/activity) plus a session/{id} template — agents get history context automatically without calling tools. 100% local, read-only, no network.
  • MCP auto-configuration — suv init claude-code and suv init cursor now automatically register the MCP server in ~/.claude.json and ~/.cursor/mcp.json. Zero extra setup for agent memory.
  • Cursor agent integration — suv init cursor installs afterShellExecution and beforeSubmitPrompt hooks into ~/.cursor/hooks.json. Captures AI agent commands with exit codes and prompts.
  • Post-install tips — all suv init commands now show actionable next steps (suv agent prompts, suv agent dashboard). Claude Code and Cursor init also hint that the agent can query history directly via MCP.
  • Enhanced suv status — shows database path, total commands recorded, detected agents, and actionable tips.

Fixed

  • Cursor executor detection — Cursor agent commands (via $CURSOR_AGENT) detected as executor_type="agent". Cursor and Antigravity checks moved before VS Code to avoid misidentification (both are VS Code forks).
  • Antigravity tagged as agent — changed from executor_type="ide" to executor_type="agent" since $ANTIGRAVITY_AGENT signals AI agent execution.
  • Unique mode sort — Ctrl+U in search now sorts by frequency (most used first) instead of recency, so common commands like git status appear at the top instead of one-off agent commands.
  • Relative date parsing — "N days ago" now works in all date inputs, fixing suv agent prompts default --after "7 days ago" which was silently returning no filter.
  • Settings list scrolling — exclusions, auto-tags, and agents lists now show scrollbars when items overflow.

Changed

  • Unified TUI styling — period selector uses pill-style highlight (bg) and left-aligned key numbers across Stats, Agent Dashboard, and Agent Stats. Executor color standardized to badge_executor, path to badge_path, session ID to primary_dim across all TUIs. All top-level TUIs show q/Esc Quit consistently.
  • Agent Stats title — renamed from "AGENT ANALYTICS" to "AGENT STATS" to match the command name.
  • Full session ID — search and dashboard detail panes show full session ID instead of truncated 8 chars.
  • Prompt shown in search — agent prompt displayed in search detail pane when available.

[0.1.5] - 2026-03-23

Added

  • Prompt Explorer — new two-screen TUI (suv agent prompts or press p in the agent dashboard) to browse agent prompts and drill into the commands they triggered. Right-side preview shows full prompt text, session, executor, path, timestamps, and success/fail stats. Supports Ctrl+Y to copy commands and s to jump to session timeline.
  • suv agent prompts CLI — direct shortcut to launch the Prompt Explorer with --after, --executor, and --here flags.
  • PostToolUseFailure hook — captures failed Claude Code commands with parsed exit codes. Run suv init claude-code to install the new hook.
  • Cursor agent integration — suv init cursor installs afterShellExecution and beforeSubmitPrompt hooks into ~/.cursor/hooks.json. Captures Cursor AI agent commands with exit codes, prompts, and session grouping.
  • Relative date parsing — date inputs now support "N days ago" (e.g. --after "7 days ago") in addition to "today", "yesterday", and "YYYY-MM-DD".
  • Executor selector in search filter — the executor field (Ctrl+F → field 5) is now an Up/Down selector showing actual executors from the database instead of free-text input.
  • Session picker improvements — live search bar (type to filter by session ID or tag), Ctrl+F filter popup with Tag/Start Date/End Date fields (matching suv search design), full session ID display (40% width), first/last command timestamps.
  • Session timeline header — shows full session ID and first/last command timestamps on two lines.
  • Alias resolution in Top Programs — suv stats resolves shell aliases in the Top Programs breakdown. Fixes #11.

Fixed

  • Dead text bug (suv search standalone) — added suv() shell function wrapper that uses print -z (zsh) / history -s (bash) to inject the selected command into the editing buffer instead of printing it as dead text. Also hardened ZLE widgets with emulate -L zsh, LBUFFER/RBUFFER, and quoted command substitutions. Fixes #6.
  • Claude Code exit codes — PostToolUse now defaults to exit code 0 (the hook only fires on success). Previously all agent commands were stored with NULL exit codes.
  • Missing recent entries in agent views — load_entries no longer truncates recent agent entries when total entry count exceeds the 10k SQL limit.
  • Session ID display — strip agent prefixes (claude-, opencode-, cursor-) before truncating for display.
  • Cursor executor detection — Cursor agent commands (via $CURSOR_AGENT) are now detected as executor_type="agent". Cursor IDE terminal check moved before VS Code to avoid misidentification (Cursor sets TERM_PROGRAM=vscode).

Changed

  • Prompt stats — None exit codes (common for agent commands) are treated as unknown, not failed. Status column shows ✔N ✘N counts instead of a misleading percentage.

[0.1.4] - 2026-03-17

Added

  • Custom agent detection — configurable [agents] section in config.toml and a new Agents tab in suv settings TUI. Define custom agent detection rules with a name, environment variable, and executor type. Custom agents are checked before built-in agents.
  • suv init opencode — OpenCode integration via plugin. Installs a tool.execute.after plugin that records every bash command OpenCode executes, with prompt capture for agent command grouping.

Fixed

  • Regex lookahead in secret redaction — the OpenAI key pattern used an unsupported negative lookahead ((?!...)), causing suv wrap and other commands to error on first use. Replaced with a compatible pattern. Fixes #9.

[0.1.3] - 2026-03-16

Fixed

  • Powerlevel10k compatibility — Ctrl+R search now works correctly with p10k instant prompt. Added ZLE display invalidation, terminal state save/restore, and queued-keystrokes guard. Fixes #6.

Added

  • Update UX — suv update now checks version before downloading, shows release notes, and displays an ASCII banner on success.

[0.1.2] - 2026-03-13

Fixed

  • Linux self-update — suv update failed with "Text file busy" (ETXTBSY) on Linux because cp cannot overwrite a running binary. Fixed by removing the old binary before copying; the kernel keeps the old inode alive for the running process.

Added

  • scripts/install.sh — Universal installer script that handles both fresh installs and updates. Auto-detects OS (Linux/macOS) and architecture (x86_64/ARM64), verifies SHA256 checksums, uses rm-before-cp to avoid the Linux "Text file busy" issue, checks version.txt to skip updates when already on latest, and only shows shell integration instructions on fresh installs.
  • Cargo install detection — suv update now detects Cargo-installed binaries (~/.cargo/bin/) and redirects users to cargo install suvadu, matching the existing Homebrew detection behavior.
  • version.txt — CI now publishes a version.txt file to downloads.appachi.tech on each release, enabling the install script to compare versions and skip unnecessary downloads.

Changed

  • Homebrew update command — suv update for Homebrew users now suggests the full brew update && brew tap AppachiTech/suvadu && brew upgrade suvadu command to ensure the tap is present and the formula index is fresh.
  • CI — Skip redundant cargo test in the macOS x86_64 cross-compile job (already tested in the ARM64 job).

[0.1.1] - 2026-03-10

Fixed

  • Allow unknown fields in config file for upgrade compatibility from older versions

[0.1.0] - 2026-03-10

A major milestone release with 75 commits since v0.0.2: new commands, secrets redaction, a comprehensive security hardening pass, architecture overhaul, and 975 tests (up from ~100).

Added

New Commands

  • suv alias — Direct shell alias management: add, remove, list, apply (write to sourceable file), and add-suggested (interactive picker from history analysis).
  • suv gc — Garbage collection: remove orphaned tags/sessions and compact the SQLite database with VACUUM.
  • suv session — Interactive session timeline TUI with a session picker, command-level detail, and scroll/filter support.
  • suv wrap — Execute and record a command without shell hooks. Designed for AI agents and scripts.

Search & Suggestions

  • Field-specific search: suv search --field cwd, --field session, --field executor to search by directory, session, or executor instead of command text.
  • Frequency-weighted suggestions in the suggest engine — commands used in more directories rank higher.
  • Improved fuzzy search ranking: length penalty prevents short substring matches from outranking exact matches; human-executed commands boosted over agent commands.
  • Help overlay (press F1 or ? in search) showing all keyboard shortcuts organized by category.
  • Responsive column layout: terminals under 80 columns show command only; 80-129 show time + command + status; 130+ show all columns. The detail pane (Tab) always shows full entry info.

Export & Import

  • JSON export format: suv export --format json alongside existing JSONL and CSV formats.
  • --json flag on applicable commands for machine-readable output.
  • Transaction rollback on import errors — partial imports no longer leave the database in an inconsistent state.

Security

  • Secrets redaction — Detects API keys, tokens, passwords, and credentials in commands and redacts them before storage. Patterns cover AWS, GitHub, Stripe, database URLs, Bearer tokens, and more.
  • Minisign signature verification for self-update: downloaded binaries are verified against a signed checksum before replacing the running binary.
  • SQLite foreign key enforcement (PRAGMA foreign_keys = ON).
  • Database file permissions restricted to owner-only (0o600).
  • Config file permissions enforcement (0o600).
  • ReDoS protection via regex::RegexBuilder::size_limit() on user-supplied patterns.
  • SQL identifier allowlisting — column names in dynamic queries are validated against a known set, preventing SQL injection via sort/filter parameters.
  • CSV formula injection prevention — cell values starting with =, +, -, @ are prefixed to neutralize spreadsheet formula execution.
  • Bounded input fields: 2,000 characters in search, 500 in settings, 256 for session IDs.
  • Session ID validation (alphanumeric, hyphens, underscores only).
  • Secure update mechanism: temp directory isolation, tar path traversal validation, mandatory checksum verification.
  • Shell-escaped hook script paths to prevent injection via directory names.
  • HTTPS enforced for all update/download URLs.

Stats & Analytics

  • suv stats --tag <name> — Filter statistics by tag for per-project analysis.
  • Stats database indexes for faster queries on large histories.
  • Hourly heatmap division-by-zero guard on empty datasets.

TUI & Display

  • Command syntax highlighting across all TUI views (search, agent dashboard, session timeline, suggest). Commands, flags, strings, variables, paths, and operators are color-coded.
  • Three-tier theme system: dark (RGB for dark terminals), light (RGB for light terminals), terminal (ANSI 16 — adapts to your color scheme). Themes hot-swap immediately in the settings UI.
  • Risk level colors centralized in theme.rs as single source of truth, replacing 15+ hardcoded Color::Rgb literals.
  • Dirty-tracking and save confirmation dialog in settings UI — unsaved changes are no longer silently lost on quit.
  • Empty state hints in agent dashboard when no commands are found.
  • Clipboard feedback message on copy.
  • Session table headers and consistent column ordering across views.

Testing

  • 975 tests total: 153 binary, 805 library, 17 integration (up from ~100).
  • Integration test suite (tests/integration.rs) covering end-to-end flows.
  • Comprehensive unit tests for: ingestion hot path, search input handlers, filter builder, stats helpers, agent UI, suggest UI, fuzzy scoring, timestamp edge cases, TUI pure-logic functions, settings flows, delete/replay /export commands, tag commands, and uninstall cleanup.

Other

  • Schema version tracking with migration framework (v1 through v4) replacing ad-hoc migration checks, with downgrade guard for forward compatibility.
  • Entry::is_agent() and ExecutorKind enum for type-safe executor classification.
  • SearchField, InitTarget, ReportFormat, SettingsTab enums replacing stringly-typed parameters.
  • 128 + signal exit code convention for signal-killed processes.
  • Detect macOS ARM architecture for correct binary downloads during self-update.
  • suv uninstall now detects and removes all installation sources (Homebrew, cargo, curl script).

Changed

Architecture

  • Module decomposition — Large monolithic files split into focused modules:
    • main.rs (1,500 lines) → commands/ directory with per-command handlers (entry, search, session, settings, stats, replay, tag, alias, wrap).
    • search.rs (2,357 lines) → search/ directory (mod, input, render, data, format, tests).
    • repository.rs (2,464 lines) → repository/ directory (mod, entries, tags, bookmarks, notes, aliases, stats, api, tests).
    • agent_ui.rs (1,620 lines) → agent_ui/ directory (mod, dashboard, stats).
    • util.rs (1,043 lines) → util/ directory (mod, terminal, format, timestamp, exclusion, highlight, file, cleanup).
    • session_ui/ — new module with picker and timeline sub-modules.
  • RepositoryApi trait — Dependency injection interface for all database operations, enabling unit tests with mock repositories.
  • SearchApp decomposition — Extracted DialogState enum, FilterState, PaginationState, ViewOptions from the monolithic search state struct.
  • SettingsTab enum — Replaced index-based tab dispatching (if tab == 2) with exhaustive enum matching.
  • Extracted Repository::init() to eliminate repeated database initialization.
  • Extracted Repository::get_tag_id_by_name() to deduplicate tag lookups.
  • Extracted build_pattern_sql helper to share SQL construction between delete and count operations.
  • FilterBuilder pattern for composable session/entry queries.
  • Eliminated in-memory entry grouping and parallel risk vector in agent UI.
  • Removed all clippy::too_many_lines suppressions via function decomposition.
  • Shared EXECUTOR_DETECTION_SCRIPT constant between zsh/bash hooks.

CI & Distribution

  • SHA-pinned all GitHub Actions for supply chain security (checkout, rust- toolchain, cache, codecov, r2-upload, gh-release).
  • Pinned cross to v0.2.5 with SHA256-verified minisign download in Linux release workflow.
  • All dependencies upgraded to latest versions.
  • Clippy lint groups enabled: pedantic, nursery, perf, complexity, style, cargo, plus unsafe_code warning.

Fixed

  • Shell hooks — Doubled braces in executor detection caused bad substitution errors on source ~/.zshrc.
  • Arrow-key navigation — Failed commands were incorrectly hidden when cycling through history with arrow keys.
  • Negative durations — Commands with clock skew or out-of-order timestamps no longer produce negative duration values (saturating arithmetic).
  • UTF-8 byte-slicing — Four locations in agent UI that sliced strings at byte boundaries instead of character boundaries, causing panics on multi-byte characters.
  • Fuzzy search threshold — Miscalculated minimum score allowed irrelevant matches to appear in results.
  • Config cache TOCTOU — Race condition between checking file mtime and reading content.
  • Timeline underflow — Empty sessions caused arithmetic underflow in timeline calculations.
  • Alias name collision — Removed arbitrary suffix limit of 99 that prevented generating unique alias names for similar commands.
  • Filter popup — Crashed or rendered incorrectly on terminals smaller than the popup dimensions.
  • Division-by-zero — Stats heatmap and percentage calculations guarded against empty datasets.
  • Thread-unsafe env access — std::env::set_var calls replaced with thread-safe alternatives.
  • LIKE escaping — Special characters (%, _, \) in search patterns are now properly escaped for SQLite LIKE queries.
  • REGEXP cache — Eliminated unwrap() on regex compilation cache that could panic on invalid patterns.
  • Nanosecond timestamps — Timestamps from tools reporting in nanoseconds are now normalized correctly.
  • Quote-aware shell chaining — Risk assessment now correctly handles &&, ||, ; inside quoted strings rather than treating them as chain operators.
  • LIMIT injection — Page size values are now validated before interpolation into SQL queries.
  • Ctrl+key fallthrough — Ctrl+key combinations no longer trigger unintended actions in search input.
  • Bookmark created_at — Timestamp now set at creation time instead of defaulting to zero.
  • Streaming export — Large exports no longer load the entire dataset into memory.
  • Display-width truncation — Uses Unicode display width so CJK and emoji characters are measured correctly instead of by byte count.
  • Atomic file writes — All configuration and data file writes use tempfile::NamedTempFile + persist() to prevent corruption on crash or power loss.
  • Eliminated all production unwrap() calls — replaced with proper error propagation or safe defaults.
  • Graceful handling of clipboard, config parsing, and JSON serialization errors.

Performance

  • Cached ProjectDirs via LazyLock — eliminated repeated filesystem lookups on every command ingestion.
  • Config mtime caching — config file is only re-parsed when the file's modification time changes.
  • Reordered early exits in the ingestion hot path — exclusion checks and validation run before any database work.
  • Streaming export — CSV/JSONL exports write row-by-row instead of collecting the entire dataset.
  • Stats query indexes — Added database indexes for the most common analytics queries.

[0.0.2] - 2025-05-20

Added

  • Post-install onboarding flow.
  • Demo GIFs in README.

Fixed

  • Deduplicate suvadu hooks in Claude Code settings.

[0.0.1] - 2025-05-18

Initial release of Suvadu — database-backed shell history for Zsh and Bash.

  • Interactive TUI search (Ctrl+R replacement) with fuzzy matching.
  • Session tracking and tagging.
  • AI agent activity monitoring with risk assessment.
  • Statistics dashboard with hourly heatmap.
  • Shell completions and man page generation.
  • Self-update mechanism.
  • Homebrew tap and curl-based installation.