v0.10.2 — Fix crates.io packaging (playwright-driver.mjs)

v0.10.2 — Hotfix

Fixed

• crates.io install broken: include_str! referenced scripts/playwright-driver.mjs outside the crate directory, causing cargo install sparrow-cli to fail with "path not found"
• Bundled the driver inside crates/sparrow-tools/scripts/ so it's included in the published package

CI

• Security audit: ignore known unmaintained/unsound dependency warnings
• Sigstore signing: marked non-blocking (binaries still downloadable)

v0.10.1 — TUI Console Redesign + CI fixes

v0.10.1 — TUI Console Redesign

TUI Console Redesign (Codex)

• Redesigned console TUI layout — cleaner, more compact, better use of space
• Improved ANSI bridge rendering
• New TUI snapshot/render scripts (PNG + SVG export)

CI Fixes

• Sandbox-linux job now green (bwrap CAP_NET_ADMIN workaround)
• Tests use LocalSandbox fallback in GHA containers

Artifacts

• TUI before/after comparison screenshots
• User-facing check screenshots

v0.10.0 « The Reasoner » — NL, Inference-Time Scaling, Swarm, Agent Framework

v0.10.0 « The Reasoner »

Natural Language (no command word needed)

• sparrow do — NL command router, just type what you want
• Bare-text front door — no prefix needed, natural language first

Reasoning Engine

• Inference-time scaling — best-of-N + Reflexion + reason_max (27B→frontier lever)
• Full Reasoning-Max integrated into default soul (no dilution)
• Fable-Grade reasoning framework — new agent/ directory with 10+ modules
• sparrow reason — live inference-time-scaling command

Swarm / Verification

• Ground-truth verifier — build/test outranks LLM judge
• Swarm verification pipeline

Agent Framework (new)

• 10+ modules: coding, communication, long-horizon, memory, reasoning-max-v2, safety, self-correction, testing, tools
• Agent config, model capability profiles, tool policies
• Evals with scoring rubric
• Master prompts (core, full, system)

UI/UX

• Clickable file/artifact links → right-side viewer in WebView
• WebView/TUI overhaul in progress
• Typography audit + proposals
• Mobile/tablet/desktop layout audit

Performance

• Boot 30% faster — skip model-discovery for read-only commands
• Streaming micro-opt: move token instead of cloning

Testing & Quality

• Provider credential-gated live smoke tests
• Router fallback chain + error-retry coverage
• Action sandbox smoke tests + Linux sandbox CI
• First-run autonomy consent notice

Security

• Landlock sandbox design
• Action sandbox hardening

Crates (workspace)

• sparrow-core v0.10.0
• sparrow-intel v0.10.0
• sparrow-providers v0.10.0
• sparrow-memory v0.10.0
• sparrow-config v0.10.0
• sparrow-tools v0.10.0
• sparrow-cli v0.10.0 (443 files, 9.2 MiB)

v0.9.3 — Security Hardening & Trust Audit

v0.9.3 — Security Hardening & Trust Audit

Security

• HardenedSandbox: Wired into exec denied-paths + SSRF pinning (S2/S3/S4 defenses)
• Trust Audit v0.9.2: Full brutally-honest audit — what's proven, partial, experimental, planned

Fixes

• Honest labels throughout README (gateways/voice/share marked 🔶 alpha)
• Hardened installers (install.sh + install.ps1)
• Fixed GitHub Action configuration
• Version coherence: package.json and docs/AUDIT.md synced to v0.9.x
• CI gates: No shipping stubs, action.yml validation, installer syntax checks

Crates

• sparrow-tools v0.9.3 (new — extracted tool implementations)
• All workspace crates bumped to v0.9.3

v0.9.2 — « The Ring » — Workspace refactor + baselines

v0.9.2 closes the ring: measure, stabilize, accelerate.

Phase 0 — Audit & Baselines

• Full monorepo audit (artifacts/audit-v092.md)
• Reproducible perf baseline: build time, binary size, startup, first paint
• Decision log (artifacts/decisions-v092.md)

Phase 1 — Stabilization

• CI stub guard: build fails if todo!() or unimplemented!() in src/
• Honest stub report (artifacts/stub-audit-v092.md)
• All formatting green

Phase 2 — Workspace extraction

Monolith split into subcrates for faster builds and cleaner boundaries:

• crates/sparrow-config/ — config, auth, hooks, humanize, permissions, sandbox
• crates/sparrow-core/ — identity, event
• crates/sparrow-memory/ — memory, FTS, redaction, symbol index
• crates/sparrow-providers/ — anthropic, ollama, openai_compat, tool markup, SSE buffer
• crates/sparrow-intel/ — intel sources

81 files, +7525 insertions. v0.9 engine with workspace-ready architecture.

v0.9.0 — « Anyone » — Radical Accessibility

Sparrow is now for everyone.

🗣️ Zero Jargon — Simple / Pro Mode

• sparrow mode simple|pro|auto — human language or expert, persisted
• Every engine Event has a FR/EN human phrase — compile-time anti-regression lock
• sparrow c'est-quoi <word> — offline glossary, no model call

🔧 sparrow fix — The Universal Fixer

• Paste an error, describe a problem, or just run sparrow fix
• Auto-diagnoses in one sentence, proposes a fix, checkpoints before acting

👋 sparrow bonjour — The Warm Welcome

• Detects your context (git conflict, uninstalled project, photo folder)
• Suggests the best action — no jargon, no flags

💡 sparrow idees — The Idea Gallery

• 50+ recipes organized by profile: child, grandparent, teacher, artisan, builder, dev
• Pick a recipe, press Enter — the recipe IS the tutorial

🖥️ Focus / Cockpit — Two Views, One Engine

• Focus (default): one column, large text, 3 buttons — Okay, Undo, Explain
• Cockpit: full 3-column with swarm lanes, config, routing
• Right-side tools panel: Preview, Diff, Terminal, Files, Tasks, Plan
• New white theme

🛡️ Safety Net

• sparrow budget 2€ — spending cap in human language
• sparrow annule — one-word undo to last checkpoint
• Contract displayed before ANY file modification

📦 One-Click Install

• First launch: zero questions, auto-detects keys, provisions free fallback
• NSIS Windows installer + macOS DMG
• sparrow launch --pro keeps the expert wizard

✅ Validated on 5 profiles

• Child (9), grandparent (68), teacher, mechanic, staff engineer
• All PASS — focus opens first, no jargon, undo always visible

51 files, +7000 lines of human surface over a stable engine.

v0.8.2 — Hotfixes

Quick fixes for v0.8.1:

• Fix: prompt caching disabled by default (400 error on non-Anthropic providers)
• Fix: missing routing_mode field in BasicRouter
• Fix: compile errors in RoutingResponse + BasicRouter
• Feat: routing manual mode (sparrow route manual/auto)
• Feat: preferred_model pinning (sparrow route set provider/model)
• Feat: WebView routing panel with model selector + manual toggle
• Feat: compact WebView cards + user sender labels
• Feat: 'no actions taken' detection for empty runs

v0.8.1 — « Honesty » (patch release, no new features)

Patch release tracking the v0.8.0 audit.

🔒 Security

• Console binds 127.0.0.1 by default, --bind flag now honored
• Clean refusal when port is already taken, --bind validation

🐛 Fixed — Tool execution (the DeepSeek bug)

• Multi tool calls no longer overwrite the first call
• Tool calls drained on finish_reason: stop instead of discarded
• Ollama native queries /api/show for live capabilities and context size

🐛 Fixed — Markup recovery

• 4 new formats recovered: JSON fences, [TOOL_CALL], DeepSeek native tags, function.arguments JSON
• Fallback parser respects string="true", no longer trims values
• Markup detection requires both open AND close structure

✅ Fixed — Approvals

• Non-interactive mode refuses cleanly instead of blocking on stdin
• Approval card uses human language via humanize_tool_action
• Web approval broker expires after hard timeout, returns deny
• WaitingForApproval status visible in cockpit

🇫🇷 Fixed — Tool calling

• French anti-narration guard: detects « je vais créer… », « laisse-moi vérifier… »
• System prompts modulated by tier (trivial/small use lean prompt)
• Curator no longer creates skills from UI status text
• History filtered at serialization: no UI status lines re-injected

📊 Fixed — Numbers

• Real provider usage replaces estimates instead of adding to them
• duration_ms in OutcomeSummary, replay uses stored duration
• DeepSeek models receive non-zero pricing, re-enabling cost comparisons
• CLI only prints 'Done.' for completed runs

🎭 Fixed — Noise

• Router message is now a single clear line instead of franglais duplicate
• Verifier lane no longer falsely shows 'run closed' without verification
• Favicon inline, no more 404 on every page load
• reasoning_content captured from single source, preventing doubled context

v0.8.0 — Projet Z: Cost Routing, Import, Self-Improving Engine

Engine: skill catalog x5, tool enforcement guard, identity fix, Distiller x5. Permissions: persistent store, AllowAlways. WebView: compact cards, syntax highlighting. TUI: keyboard hints. Import: Claude Code, Codex, OpenCode. Update: notifications. Soul: CLAUDE FABLE 5 + REFLEXION-MAX V2 fusion. main.rs: 4991 to 1509 lines.

v0.7.1 — Security audit + package cleanup

Security\n- Removed launch scripts containing local paths from published crate\n- Excluded ROADMAP.md from crates.io package\n- Added Cargo.toml exclude globs for dev-only scripts\n\n### Changed\n- v0.7.0 → v0.7.1 (package-level fix only, no code changes)\n\nFull v0.7.0 features: cost routing, sparrow import, self-improving engine, REFLEXION-MAX PROTOCOL V2