v3.18.0

Fixed

• install.sh Windows split-brain (#269): on MINGW/MSYS/CYGWIN the installer now targets %APPDATA%/devin (where the CLI actually reads) instead of ~/.config/devin; audit.py prefers %APPDATA% on Windows and warns when both dirs exist. Previous bash installs had written a shadow config that drift checks validated as "live".
• Exporter secret masking and pre-push gate (#268): export.sh/export.ps1 now mask by key-name (token|secret|api[_-]?key|password|auth|credential|private|pat, e.g. GITHUB_PAT, Authorization, client_secret) and embedded URL userinfo; TOML masking tracks multiline openers ([, """, ''') so continuation lines no longer leak; the pre-push audit.py/pytest gate now runs inside the bundle root (it previously validated the caller's cwd, where audit.py walks ., silently broken); git add -A replaced with export-owned paths; paths passed via sys.argv instead of shell interpolation; bundle-only/stale export entries now reported.
• Installer/skill robustness (#264): unreadable-file guards, ppisp subprocess timeout, empty extension directory skip in install.sh (parity added to install.ps1 in #269).
• comfyui-operator hardening (#263): loopback-only server URL and download path sanitization.
• Typed-error hardening (#262, #255, #254): plan_check, ASR bootstrap, sc_bench, custom presets, laya-compactor pairing/reduction accounting.
• asr Windows discovery (#256): whisper binary/model discovery fixed on Windows.
• Unexpanded placeholders (#272): {{VALIDATED_CLI_VERSION}} in ask-bundle (never substituted by any pipeline) now references validated_cli_version; data/bundle-identity.json filled with real values (Leostruka/devin-bundle, 2026); blender-operator POSIX ps call gained timeout=15 parity.
• Unbounded subprocess calls (#274): timeout=15 added to the PowerShell CIM query in extensions/computer-use/probes/probe_spawnpty.py; timeout=60 added to CLI subprocess calls in extensions/media-tools/tests/{test_cli,test_export}.py.
• Phantom recipe tool (#275): data/recipes.json explore listed glob (premise corrected in #284: glob is a real tool, but recipes should not demand mode-absent tools); it capped the recipe tool score at 0.75 in context-pressure.py. Removed; regression test validates every recipe tool against the canonical CLI tool set.
• False architecture boundary + missing enforcement (#278): .devin/ARCHITECTURE_MANIFEST.md claimed hook scripts never import from extensions/, but laya-guard.py intentionally does (dynamic, fail-open); manifest corrected to "no third-party pip deps" with the laya exception documented; new test_scripts_import_boundary test enforces stdlib/sibling/exception imports across scripts/*.py.
• POSIX reap verification flake (#280): _reap_posix_group group-absence deadline widened 5s -> 15s; SIGKILL-to-launchd-reap latency on loaded macOS CI runners exceeded 5s, producing transient false tree_cleanup results (test_spawn_timeout_cleans_tree failed 3x).
• Phantom tool in skill frontmatter (#282): observability-quality allowed-tools listed glob (premise corrected in #284) corrected to find_file_by_name; new test_skill_allowed_tools_exist enforces canonical tool names across all SKILL.md frontmatter.
• glob tool misclassified as phantom (#284): glob is a real CLI tool (validate-tool-args checker + TOOLS-MAP); the session toolset is mode-dependent and not canonical. #282's premise corrected: observability-quality restored to glob + find_file_by_name; VALID_TOOLS expanded to the canonical 28-tool set; allowed-tools test now covers YAML-list form and agents/ + .devin/agents/ profiles.
• Orphaned extension requirements (#287): playwright (craft-bridge), opencv-python-headless and piexif (image-forensics) were declared but never imported; ppisp-prep gained the missing requirements.txt (numpy, pillow) for its hard deps; new test_extension_declared_deps_used enforces declared->used per extension.
• agent-reach wrapper desync from upstream CLI: wrapper forwarded channels as subcommands (agent-reach youtube ...), a grammar agent-reach 0.1.0 never shipped; every documented call exited 2 and doctor reported degraded probing a nonexistent --version. Wrapper now translates <channel>[.<command>] to get <target> --json, doctor proxies doctor --json per-channel status, whitelist narrowed to the real upstream index (youtube, rss; auth:none), and SKILL.md/manifest purpose synced (previous docs claimed phantom github/bilibili/v2ex/web channels).

Added

• test_extension_used_deps_declared (#288): the reverse direction of the R20 check is now enforced — module-scope third-party imports in extensions/ must be covered by that extension's requirements*.txt; AST-based (function-scope lazy imports exempt), extensions-internal module roots exempt, documented bpy/mathutils host-provided allowlist for Blender's embedded interpreter; the R20 orphan check was upgraded from regex to AST.
• agentic-rag vectorless/tree-index retrieval (#266): architecture selector gained long structured docs -> PageIndex-style with honest costs (LLM call per hop, index build time, stale-tree versioning); gate/escalation unchanged.
• brag portfolio-site beat order (#267): storyboard template for portfolio/personal-site inputs (hero -> work -> recognition -> process -> values -> CTA) from real-site reel re-analysis.
• stuck-detect deterministic advisory (#260): token-economizer subset hook.
• Ponytail minimal-diff checklist (#258): Goldfish zoom-out review step.
• EGM runtime rollout (#253): JEV routing, OpenUI runtime, camera blocking, BitNet/offloading adapters, TPU held-out scaffolding + 12-issue audit hardening.

Changed

• MODEL-GUIDE: Avalon AMD MoE fork evaluated-not-adopted with 4-question benchmark-reading checklist (#265); LFM3 local-audio family entry (#259).
• dispatching-parallel-agents (#277): DeepSeek super-agent eval evidence added; task-queue interruption inverts the sub-delegation bias (neutral prompt 0/5, delegate-first 4/5, median 20.4->13.8 min) plus Cooper Bench co-edit cost (~30% lower success).
• audit.py check 32 (#281): [Unreleased] merge coverage now enforced; merge PRs since the released tag without #NNN refs in [Unreleased] fail the audit (docs(changelog) PRs exempt).
• audit.py pycache check robustness (#285): check 11 git ls-files pathspec replaced by Python-side '__pycache__' in path filter, matching tracked caches at any depth.

Documentation

• Source analyses: NCM coarse-to-fine (#261), FreeMyTv Claude-Code-on-TV (#257), batch sources incl. DeltaDB/TradingAgents/Kimi K3/Openjutsu (#265), TSF Lesson 1 (#271), GG Coder demo (#276), all notes-only, no portable implementation; Nx-style lib boundaries applied internally as a real enforcement fix (#278); R8 issue answers (#270); review-pattern capture and drift reconciliation (#252).

v3.17.0 - 10 ported discipline/security skills

v3.17.0

Ported 10 skills from a parallel Devin install, adapted to bundle conventions (skill_count 91 -> 101):

• powershell-safe, shell-safe: shell footgun traps (redirect BOM/truncation, empty-var rm, pipe exit codes)
• edge-cases: input/state/failure-mode enumeration before coding
• dead-wire: dead-feature audit (buttons, events, flags with no consumer)
• github-actions: CI security (SHA pinning, pull_request_target, OIDC, cache poisoning)
• webhooks: receiver discipline (signature verification, idempotency, fast-ack)
• typescript-node: strict-mode/async/npm footguns
• binary-triage: fast static evidence from closed binaries pre-Ghidra
• ghidra: validated headless workflow (Thumb/TMode, seeds, firmware)
• dma: stdlib-only scoped recon/probe toolkit (scope.md + 10 tools, gated invasive classes)

Issue

Closes #250 (milestone v3.17.0)

Verification

• audit.py: 32 checks, zero errors/warnings
• validate-skill-format.py: 251/251
• CI: 12/12 checks green (PR #251)

v3.16.0

Changed

• Devin CLI 3000.11.234 alignment: validated_cli_version bumped; new capability table (--verbosity/DEVIN_VERBOSITY, hook session_id/prompt_id, updatedInput, keymap validation, devin desktop); agent.verbosity: 3 note corrected.
• agentic-rag production guardrails: architecture selector, 3-context separation (conversation / user memory / documents), iron rules against silent rerank/gate fallback and unbounded streams, hit@k/MRR/p50/p95 metrics in Gate G5.
• ai-tools post-training playbook: evidence-backed method selector, 4-gate route (verifier -> baseline -> distillation -> RLVR/GRPO/DAPO), 6 iron rules, 8 arXiv references.

Closes #247, closes #248. Milestone: v3.16.0.

v3.15.0

Added

• brag-slim skill (#244): single-file /brag variant vendored from latent-spaces/brag v0.5.0. No bundled assets; builds the launch video with tools already on the machine and supports website-URL input with prompt-injection hardening (fetched content treated as data, never instructions). Full skills/brag/ synced to upstream 0.5.0: Opus 5.5 dispatch to /brag-slim, bundled slim.md, refreshed cue-analysis deps. Manifest, README and TOOLS-MAP synced to 91 skills.

Fixed

• Stale composition-skill references in vendored /brag (#245): the delegation pointed at HeyGen domain skills absent from the bundle. Composition guidance now uses skills/hyperframes + extensions/hyperframes/wrapper.py for the CLI contract and skills/animation-direction for motion and beat-to-visual mapping. HeyGen skills stay optional when their plugin is installed; audio-reactive extraction documents a local ffmpeg/librosa fallback, and .claude paths were replaced with Devin install paths.

v3.14.0

Added

• drift-report skill (#242): port of egm-drift-report (google/elephant-goldfish-model, Apache-2.0). Detects design-doc-vs-code drift: missing features, incomplete models, logic shortcuts, architectural divergence. Report-only output <YYYY-MM-DD>_DRIFT_REPORT.md; complements spec-consistency.py (plan-file internals) with doc-to-code fidelity.

Fixed

• Audit drift warnings (#241): render_norm() now normalizes pythonw -> python (false-positive hooks live-vs-render after the pythonw renderer in #232); manifest.json export_hash re-synced for 10 scripts + 2 agents stale since #234/#236/#240; README version badge synced with manifest.

v3.13.0

Added

• agentic-rag skill (#237): implementation playbook for evidence-gated agentic RAG products (retrieve -> decision-gate -> answer-or-escalate). Multimodal per-page index (text chunks + page PNG), decision-model answerability thresholds in DB, human-review ticket loop with save-to-knowledge re-indexing. Stack-agnostic slots; decision-model slot is free-first with Laya multilingual (Apache-2.0, self-host) as reference, Kev/Jeeves/Strands Decider/open-weight d1-3B as swaps, Jev/hosted d1 as paid options.

Fixed

• Windows console popups (#238): every console-subsystem child spawned from the console-less runtime allocated a visible conhost window. Hook renderer emits pythonw on Windows (#232); all subprocess call sites in hook scripts (#234) and across extensions (#236, 29+ sites) pass CREATE_NO_WINDOW via getattr fallback (POSIX no-op).
• CU toast identity (#239): Windows notifications now show Cu-Talk instead of devin-computer-use; MCP protocol name unchanged.

v3.12.0

Added

• External-repo integration suite (Stackademic top-10 analysis, #224, #226, #227, #228, #229): 10 repos evaluated against the bundle; 5 landed as skills/wrappers, 4 already covered (brag, grilling/tdd/implement/triage, rea-ops, scrape-tools), 1 rejected (VoiceStudio: AGPL + desktop GUI).
• frontend-design skill (#224): port of anthropics/skills frontend-design (Apache-2.0); aesthetic direction up front, pairs with impeccable.
• career-ops skill (#229): port of santifer/career-ops workflow modes (MIT); public-board scan, fit scoring, CV tailoring, tracker under .devin/scratch/career-ops/; agent never submits applications.
• hyperframes extension + skill (#226): governed wrapper (extensions/hyperframes/wrapper.py) for npx hyperframes HTML-to-MP4 rendering, whitelisted doctor/init/render; HTML-native alternative to remotion-render.
• agent-reach extension + skill (#227): governed wrapper for the agent-reach JSON CLI, limited to credential-free channels (github, youtube, bilibili, v2ex, web); cookie channels rejected per bundle credential policy.
• tts skill + voicebox wrapper (#228): local TTS via the voicebox FastAPI backend (127.0.0.1:17493, urllib-only client); completes the local narration pipeline tts -> asr -> animation-direction.
• Contract tests for the three wrappers (25 tests, no network).

v3.11.1

Fixed

• manifest.json version sync: bumped 3.10.0 -> 3.11.1 so the release workflow's version gate passes. The v3.11.0 tag shipped without the manifest bump and its release run failed at Verify tag ancestry and version sync; tags v* are immutable, so the fix lands as this patch release. All 3.11.0 content ships unchanged in v3.11.1.

v3.11.0 - Devin CLI 3000.11.228 alignment

Changes

• fail_closed: true on PreToolUse guards: a failing gate hook now blocks the action instead of letting it through (verified empirically).
• read_config_from.cloud: false: Devin Cloud MCP servers and personal plugins stay out of the validated surface; organization-required plugins still load.
• validated_cli_version -> 3000.11.228 (installed 053434014dbd); compatibility doc gains the 3000.11.205/228 capability decision table; TOOLS-MAP updated (/rename, /confetti, /hooks, devin models, new hook fields); /loop + agent.model_mixture removal noted (replaced by /goal).
• keymap.readbox_input.queue_message rebound to ctrl+4.

Closes #219, #220.

v3.10.0

Added

• agent-rts LORE.md: DevinCraft, A Cronica da Oficina (extensions/agent-rts/LORE.md, #213): 7-section pt-BR tale mapping the 11 canonical buildings, 15 hero classes, fog, contested tiles, Wonder Age, War Council and the player beyond the mist; extension-local ARCHITECTURE_MANIFEST.md for the architecture gate.
• agent-rts 50-loop WC3 polish pass (#214): combat (hit flash, death dissolve+pool, damage vignette, camera shake, under-attack ping+horn, hero-fallen alert, +xp floaters), control (Tab subgroups, contextual r-click, ESC disarm, control groups, formation spiral, hover ring), world (soft fog + reveal peel, void edge, instanced doodads, cloud shadows, critters, contested decals, gold piles, day/night with sun/moon/dial, night windows), meta (rally line+flag, waypoint posts, production bars, decision beacon, quest+age banners, per-class barks, ambient pad, favicon). Matrix: 51 DONE, 1 GAPPED, 4 BLOCKED-BY-CANON.

Fixed

• agent-rts fog of war leaks + army vision (#208, #209): marks/plots/gold/decals/smoke/dust/sparks/server FX/meteor trails no longer render inside unexplored shroud; puff(x,y,z) honored its z argument (all smoke compressed to the north edge); own troops never vanish and carry a sight halo via a separate fog overlay plane repainted per unit-tile change; path trail counter-rotated so waypoints anchor to the map instead of spinning with facing.
• agent-rts camera rebuild (#210): pan/zoom write directly to the camera (no ~150ms spring lag), rotation only via Insert/Delete/MMB-drag, RMB never rotates (fixes smart-order collision), rotated-pan sign error fixed.
• agent-rts HUD regression (#211): banner skin scoped to #banner (was stacking portrait/feed/roster/toasts at top-center), inventory runes render readable values, feed dedups consecutive repeats, camera rescue snaps instead of gliding from the void, window.__rts debug handle.
• UTF-8 decoding on Windows pipes (#212): collectors._run and check-ai-signature decode subprocess output as UTF-8 with replace (was cp1252: ticket mojibake frágeis, hook crash on UTF-8 diffs).
• computer-use record flake (#118): _record accepts injected _now/_sleep; pacing tests run on a deterministic FakeClock instead of racing a loaded CI scheduler.