Releases
Every release, with its date.
126 releases across the two public repos, newest first. The latest engine build is v5.0.0-beta.51. Snapshot taken ; the list regenerates on every site build.
gald3r_core — the engine and every signed binary
The Go rewrite. Each release ships the signed gald3r CLI, the MCP server, gald3r_ide, Longship, Cartograph, and the Valkyrie connector for Windows, macOS, and Linux.
Throne, gald3r_ide and the CLI now ship together on one tag. Highlights Throne ships with gald3r now. This is the first release where the Throne desktop app and the new gald3r_ide editor are built and published on the same tag as the CLI, at the same version. The separate Throne 0.x line is retired; install Throne…
Release notes on GitHubThis cut ships gald3r_ide, a standalone code editor built on the same editor core as Throne, and puts every AI edit in front of you as a diff before it touches disk. It also fixes the four problems reported from a real workplace install this week. gald3r_ide, a new binary One screen: file tree, tabs, editor, a…
Release notes on GitHubCompany admins run their own team, and the IDE shows you every AI edit before it touches a file. See the change before it lands In the Throne IDE, a proposed edit to the file you have open appears as a diff. Accept or reject each hunk, or all at once; nothing is written until you press Apply, and only the hunks you…
Release notes on GitHubYour project agents can now talk, and you can put all of them in one room. Talk to your project agents The per-project chat answers through a model when one is available: it reads the project first (search, board, status) and asks a local runtime such as Ollama or LM Studio to answer over that context. Name a…
Release notes on GitHubNotes Throne starts showing you the whole workspace, and the engine stops accepting work nobody can verify. Throne: see the workspace, not just the project A Workspace graph opens the Workspace page: every project in your workspace as a node with its name and project type, a lit indicator when that project's Valkyrie…
Release notes on GitHubYour project's agent can now address another project's agent on the event wire — delivery still waits on a server catalog deploy. Ask another project Same-machine asks still write the local mailbox, as before. gald3r project-agent ask --to-project <id> … queues a remote question and flushes the outbox immediately.…
Release notes on GitHubYour repo can answer you now. Talk to it from the terminal or a slash command — same brain as Throne — instead of opening a second editor window on the board. Talk to the project gald3r project-agent status, ask, and chat — or @g-project-agent Same engine Throne and the resident connector already use Set…
Release notes on GitHubThis beta closes a serious board-integrity defect class, removes the biggest throughput bottleneck in autonomous runs, and cleans internal-only content out of the installed payload. Your completed work stays completed. A multi-day defect class that could silently revert already-resolved tasks and bugs during…
Release notes on GitHubYour plan, your spend, your call. This release rebuilds autopilot's execution model around billing trust. The orchestration loop now runs its workers inside the driving session by default — billed to the subscription you already pay for, never a separate metered surface. Every launch declares each role's billing…
Release notes on GitHubTheme: the loop grows judgment. Hours after the last beta, this one lands the reliability batch that a day of live field runs asked for — plus the first slice of the biggest architectural change the autopilot has had. A pre-review gate with zero model cost. Before any reviewer is spawned, a deterministic gate checks…
Release notes on GitHubTheme: an autopilot you can leave running. This beta is mostly about the autopilot loop finishing what it starts. Two days of monitored multi-lane runs surfaced a family of boundary failures — work lost between runs, verified work demoted because of a merge hiccup, fixes dying at rebase for reasons that were never…
Release notes on GitHubThis release closes the most expensive failure mode reported from live autonomous runs: work that fails review over and over, silently burning budget with no human ever told. Autonomous runs now know when to stop and ask. An item that fails review three times is parked for human attention instead of being sent back…
Release notes on GitHubThis release closes the second wave of field reports from live multi-agent runs — every fix below was reproduced from a real field failure and shipped with a regression test. Autonomous runs can no longer hang, lie, or burn silently. A vendor CLI that stops producing output is now detected within minutes and its…
Release notes on GitHubThis release hardens the autonomous pipeline based on the first wave of beta.35 field reports — every fix below came from watching real multi-agent runs at three field sites. Review gate integrity. Implemented work is now held off your main branch until the independent review actually passes. A failed review leaves…
Release notes on GitHub5.0.0-beta.35 -- the tight-ship build. Every leak, lie, and blind spot the field found in the loop this week is closed -- eleven verified fixes, shipped the morning after the reports came in. The dispatcher obeys its own math. Excluded or ineligible work can no longer be spawned anyway: the scheduler consumes exactly…
Release notes on GitHub5.0.0-beta.34 -- the safe-hands build. Your money, your work, and your results are all handled with surer hands this release. Idle never burns. An autopilot run with nothing dispatchable no longer pays a coordinator to rediscover an empty queue -- the loop skips the spend outright and halts cleanly under a new…
Release notes on GitHubThe honest-numbers build. Every number the loop reports is now true, and every dollar it spends is governed. The cost governor is real. The loop now reads each agent's actual reported spend (the data was always there -- now it's parsed), so the final cost figure is truthful and the spending ceiling actually stops a…
Release notes on GitHubThe recovery build. Nothing is wasted and nothing is lost: the loop stops re-doing finished work, recovers a crashed agent's train of thought, cleans up after every failure, and its agents can now talk to each other. Already-done work stays done. Items whose changes already landed on your main branch are detected…
Release notes on GitHubThe fortress build. Whole classes of annoyance become impossible: no more focus-stealing windows, no more health checks that lie, and a project status you can read at a glance. Focus theft is over. Every child process gald3r spawns — during runs, tests, or tooling — now launches through one shared hidden-spawn…
Release notes on GitHubThe discipline build. The autonomous loop's git work gains branch discipline, every lane's pulse is visible live, and the run ledger is now forensically complete. Branch discipline, opt-in. Two new modes for the autonomous loop, both off by default. *Session branches*: the loop's own bookkeeping commits land on a…
Release notes on GitHubThe trust build. A review's verdict is law, the docs site is professional grade, and nothing internal leaks to the outside. Review verdicts always apply. A reviewer's written verdict can never again be suppressed because its prose resembled an error message; and an item whose reviews keep failing to conclude parks…
Release notes on GitHubThe resilience build. The loop now survives its own worst moments — and leaves a complete forensic trail when anything goes wrong. One stuck task can never crash a run again. The crash class where a requeued work item could take down the whole loop mid-run is fixed at three layers: the root ordering bug, graceful…
Release notes on GitHubThe supervision build. This release is about one promise: when the autopilot loop is running, it is either shipping work or telling you exactly why not — never silently spinning. Reviews now finish everything. The review phase can verify bug fixes as well as tasks, so a queue of completed-but-unverified bug work…
Release notes on GitHubThe throughput build. beta.26 makes the loop spend its budget on work an agent can actually finish — and hand everything else to you, once, instead of retrying it forever. The loop stops re-attempting items a human has to unblock. A failed review whose remaining acceptance criteria are human-gated parks the item in…
Release notes on GitHubThe stability build. beta.25 consolidates two weeks of rapid field-driven fixes into one hardened release: the loop tells the truth about its own state, upgrades are safe, and budgets finally mean what they say. **Budget you can explain in one sentence: it is the number of tasks and bugs the loop will attempt to…
Release notes on GitHubAutopilot now reviews each finished item in the same loop instead of waiting for a later batch pass. Completed task files survive harvest. Leftover scratch files and local scheduler or vault-ingest state no longer stop a run. Worktrees keep the controller board database. Child checkouts no longer rebase shared board…
Release notes on GitHubAutopilot now runs implementers in parallel by default. Use --no-code-swarm if you want the old one-at-a-time coding path. Sibling count defaults to 5 (recommended 3–8; higher values are allowed and warn). --budget is an attempt cap; 0 means unlimited. Vault-first research is on: agents search your local…
Release notes on GitHubg-go, g-go-code, and g-go-review now stay on the model of the IDE you launched them from. They no longer switch to a pinned model from project config. Autopilot (gald3r autopilot loop / @g-go-go) requires --provider and --model on the call (or the per-role flags) and prints which model each role will use before it…
Release notes on GitHubThis beta is a field fix so work and world_tree hosts can run autopilot without the mid-loop halt from the last backlog run. Board-file dirt written during a loop no longer blocks harvest or the next worktree. Autopilot now housekeeps those allowlisted writes between iterations. A Critical bug that is only a title…
Release notes on GitHubAutopilot (gald3r autopilot loop / @g-go-go) now runs multiple implementers at once by default. Use --no-code-swarm when you need one-at-a-time (debugging, or a host that cannot run several coding agents together). Merge-back stays serialized so parallel worktrees do not wedge the git lock.
Release notes on GitHubField fixes from the beta.17 Windows/Cursor run, for testers who already installed beta.18. Conductor vs parallel work. @g-go-go / autopilot loop stays sequential (that honesty shipped in beta.18). This cut makes the sequential path survivable: the liveness heartbeat can see bucket output, silent buckets are killed…
Release notes on GitHubAutopilot loop docs now match the binary: @g-go-go runs one implementer at a time. Parallel coding is @g-go --swarm, not a hidden loop flag. Field reports (gald3r gfix) warn in the draft itself when GitHub CLI cannot auto-file, fill watchdog evidence from the last captured kill when you type the class by hand, and…
Release notes on GitHubThis build is the field follow-up to 5.0.0-beta.16. Autopilot now treats a finished no-commit turn as real work instead of an idle failure, so a correct "nothing to commit" triage no longer trips the capacity breaker and kills a run that still has queued work. Claims held by buckets that die at run-end are released…
Release notes on GitHubSpawned agents now receive their complete instructions — every time, at any size. This release closes the last known gap in the orchestration pipeline: agent briefs are now delivered whole, never truncated. Complete-brief delivery. Large agent briefs (coordinator, implementer, and reviewer) are written to a file…
Release notes on GitHubThis release delivers the first real autonomous throughput on every supported configuration — and gives users a one-command way to report anything that still goes wrong. Worker agents actually work now. The agent CLI is invoked correctly for non-interactive use (the missing print-mode and trust flags explain both the…
Release notes on GitHubThis release reorganizes every project around a cleaner structure and adds three new ways to capture and commit to work — plus first-supporter integration for a new local AI engine. A cleaner project layout. SDLC working files (tasks, bugs, plans, features, PRDs, ideas, and the new artifact types below) now live…
Release notes on GitHubThis release restores the autonomous pipeline's full operating knowledge and makes its safety checks real. The autopilot coordinator got its complete instructions back. The condensed internal briefing that made autonomous coordinators unreliable at delegating work has been replaced with the full, field-proven version…
Release notes on GitHubgald3r 5.0.0-beta.1 — the first release of the Go rewrite. gald3r has been rewritten from the ground up in Go: ~100ms startup (down from 8–11s), a single fast binary set, signed and notarized on Windows, macOS (Apple Silicon + Intel), and Linux. Install Windows — download gald3r-windows-x86_64.msi, run it, open a new…
Release notes on GitHub- v4.0.0-beta.1pre-release
gald3r 4.0.0-beta.1
First 4.0 prerelease: a signed, self-contained Windows binary. What's here gald3r.exe — Authenticode-signed as *Gald3r Labs LLC*, timestamp countersigned gald3r.exe.sha256 — verify your download before running it Verify (Get-AuthenticodeSignature .\gald3r.exe).Status # -> Valid (Get-FileHash .\gald3r.exe…
Release notes on GitHub
v5.x
gald3r — templates, skills, and the wiki
The public templates repo: the CRASH catalog (commands, rules, agents, skills, hooks) that gald3r platform install projects into your AI coding tool, and its release history back to v1.0.0.
gald3r v5.0.0-beta.51
Throne, gald3r_ide and the CLI now ship together on one tag. Highlights Throne ships with gald3r now. This is the first release where the Throne desktop app and the new gald3r_ide editor are built and published on the same tag as the CLI, at the same version. The separate Throne 0.x line is retired; install Throne…
Release notes on GitHubgald3r v5.0.0-beta.50
This cut ships gald3r_ide, a standalone code editor built on the same editor core as Throne, and puts every AI edit in front of you as a diff before it touches disk. It also fixes the four problems reported from a real workplace install this week. gald3r_ide, a new binary One screen: file tree, tabs, editor, a…
Release notes on GitHubgald3r v5.0.0-beta.49
Company admins run their own team, and the IDE shows you every AI edit before it touches a file. See the change before it lands In the Throne IDE, a proposed edit to the file you have open appears as a diff. Accept or reject each hunk, or all at once; nothing is written until you press Apply, and only the hunks you…
Release notes on GitHubgald3r v5.0.0-beta.48
Your project agents can now talk, and you can put all of them in one room. Talk to your project agents The per-project chat answers through a model when one is available: it reads the project first (search, board, status) and asks a local runtime such as Ollama or LM Studio to answer over that context. Name a…
Release notes on GitHubgald3r v5.0.0-beta.47
Throne starts showing you the whole workspace, and the engine stops accepting work nobody can verify. Throne: see the workspace, not just the project A Workspace graph opens the Workspace page: every project in your workspace as a node with its name and project type, a lit indicator when that project's Valkyrie…
Release notes on GitHubgald3r v5.0.0-beta.46
Your project's agent can now address another project's agent on the event wire — delivery still waits on a server catalog deploy. Ask another project Same-machine asks still write the local mailbox, as before. gald3r project-agent ask --to-project <id> … queues a remote question and flushes the outbox immediately.…
Release notes on GitHubgald3r v5.0.0-beta.45
Your repo can answer you now. Talk to it from the terminal or a slash command — same brain as Throne — instead of opening a second editor window on the board. Talk to the project gald3r project-agent status, ask, and chat — or @g-project-agent Same engine Throne and the resident connector already use Set…
Release notes on GitHubgald3r v5.0.0-beta.44
Your completed work stays completed. This release closes a defect class that could silently revert already-resolved tasks and bugs during autonomous merges — closed structurally, at every layer: agent worktrees no longer see board records at all, commit safety fails closed whenever a merge or conflicted state is in…
Release notes on GitHubgald3r v5.0.0-beta.43
The autopilot finishes what it starts. Completed work now merges itself — an automatic conflict-resolution ladder lands every finished branch, interrupted work is rescue-committed before anything can be lost, stranded branches get garbage-collected, and blocked items unblock the instant their blockers complete. A…
Release notes on GitHubgald3r v5.0.0-beta.42
Your plan, your spend, your call. Autopilot's execution model is rebuilt around billing trust: the loop now runs its workers inside the driving session by default — billed to the subscription you already pay for, never a surprise metered surface. Every launch declares each role's billing surface in color (green: your…
Release notes on GitHubgald3r v5.0.0-beta.41
The field speaks, the same day it ships. Every fix in this cut traces to real project reports filed against yesterday's build. Deleting a bug record can no longer wedge a project half-done — the operation is transactional on every layout, recovers cleanly from any earlier stuck attempt, and its errors now tell the…
Release notes on GitHubgald3r v5.0.0-beta.40
The loop grows judgment. Cut hours after beta.39, this build lands the reliability batch a day of live multi-lane runs demanded: a zero-model-cost pre-review gate bounces bad handoffs in milliseconds (with the release-note requirement computed from the diff itself); a harvested implementation only earns credit once…
Release notes on GitHubgald3r v5.0.0-beta.39
An autopilot you can leave running. Two days of monitored multi-lane runs surfaced a family of boundary failures -- work lost between runs, verified work demoted because a merge could not land, bug fixes dying at rebase for reasons that were never content conflicts -- and each is closed: interrupted implementations…
Release notes on GitHubgald3r v5.0.0-beta.38
Autonomous runs now know when to stop and ask. An item that fails review three times is parked for human attention instead of being retried forever; a reviewer that gives up mid-review counts as a failed review; and database records with no backing file can no longer be offered to reviewers -- ending the loop where…
Release notes on GitHubgald3r v5.0.0-beta.37
The truth-telling build. Closes the second wave of field reports from live multi-agent runs. Autonomous runs can no longer hang, lie, or burn silently: a vendor CLI that stops producing output is detected within minutes and cleaned up instead of holding a claim for hours, every run — including crashes — ends with a…
Release notes on GitHubgald3r v5.0.0-beta.36
The earned-merge build. Implemented work now stays off your main branch until the independent review actually passes — a failed review leaves main untouched and keeps the work aside for the retry, and loaded project constraints are enforced as a hard pre-merge gate. The autopilot recognizes when an open bug was…
Release notes on GitHubgald3r v5.0.0-beta.35
The tight-ship build. Eleven verified fixes shipped the morning after the field reports arrived: the dispatcher obeys its own eligibility math; failures can never masquerade as successes (no-commit branches, narration, overlay-only evidence, and admits-unchanged passes all guarded); hangs caught in minutes via the…
Release notes on GitHubgald3r v5.0.0-beta.34
The safe-hands build. Autopilot idle never burns money (skip + idle-iteration limit); legitimate no-code-change fixes count as successes (no false failures or capacity downgrades); crashed sessions' work lands on dedicated rescue branches, never the working branch; cross-project inbox rebuilds union files with the…
Release notes on GitHubgald3r v5.0.0-beta.33
The honest-numbers build: every number true, every dollar governed. Real cost governance: actual agent spend parsed and accumulated; the spending ceiling works; budgets are hard (retries and resumes count). No self-deadlocks: the loop's database files are guaranteed ignored everywhere, and automated commits can never…
Release notes on GitHubgald3r v5.0.0-beta.31
The fortress build: whole classes of annoyance become impossible. Focus theft is over: every child process spawns through one shared hidden-spawn mechanism, enforced by an automated gate so the fix can never erode. gald3r status reorganized into grouped sections -- tasks, bugs, your board, suggested next actions…
Release notes on GitHubgald3r v5.0.0-beta.30
The discipline build: branch discipline for the loop's git work, live lane visibility, and a forensically complete ledger. Two opt-in branch modes: session branches collect the loop's own bookkeeping commits into one reviewable unit; branch-in-place runs solo sequential work on a feature branch in your checkout with…
Release notes on GitHubgald3r v5.0.0-beta.29
The trust build: verdicts are law, the docs are professional, nothing internal leaks. A reviewer's written verdict can never be suppressed because its prose resembled an error message; items whose reviews never conclude park for human attention after three attempts. The documentation site rebuilt: dedicated Commands…
Release notes on GitHubgald3r v5.0.0-beta.28
The resilience build -- plus a thinking toolkit and two navigation aids. One stuck task can never crash a run: root fix, graceful refusal, and a recovery net around the scheduler. Budget exhaustion finishes in-flight work (merge + review) before ending; leftovers report honestly in one line. Blocked work stays…
Release notes on GitHubgald3r v5.0.0-beta.27
The supervision build: a running loop is either shipping work or telling you exactly why not -- never silently spinning. Reviews finish everything: bug fixes get fresh-eyes verdict turns too, so completed-but-unverified work drains instead of accumulating (gald3r go-bug-review). Fresh work outranks repeat offenders:…
Release notes on GitHubgald3r v5.0.0-beta.26
The throughput build: the loop spends its budget on work an agent can actually finish, and hands everything else to you -- once. Human-gated items park in your attention queue after one attempt instead of being re-implemented forever; every item caps at three attempts per run with a named reason. The picker never…
Release notes on GitHubgald3r v5.0.0-beta.25
The stability build: state reconciliation, safe upgrades, and budgets that mean what they say. Budget in one sentence: the number of tasks and bugs the loop will attempt to handle. Reviews never consume budget, and completed work gets reviewed and finalized before anything new starts. The board never lies: failed or…
Release notes on GitHubgald3r v5.0.0-beta.24
Autopilot reviews each finished item in the same loop. Harvest keeps completed task files. Autopilot now reviews each finished item in the same loop instead of waiting for a later batch pass. Completed task files survive harvest. Leftover scratch files and local scheduler or vault-ingest state no longer stop a run.…
Release notes on GitHubgald3r v5.0.0-beta.23
Autopilot runs implementers in parallel. Vault-first research is on. Autopilot runs implementers in parallel by default. Use --no-code-swarm for one-at-a-time. Sibling count defaults to 5 (recommended 3–8). --budget is an attempt cap; 0 means unlimited. Agents search the local knowledgebase before the internet.…
Release notes on GitHubgald3r v5.0.0-beta.22
g-go stays on the model you launched from. Autopilot names its models up front. g-go / g-go-code / g-go-review inherit the IDE you are in. They no longer switch to a pinned model from project config. Autopilot (gald3r autopilot loop / @g-go-go) requires --provider and --model (or the per-role flags) and prints which…
Release notes on GitHubgald3r v5.0.0-beta.21
Field fix so a live autopilot loop can finish and harvest. Allowlisted board-file writes mid-loop no longer block harvest or the next worktree. A title-only Critical bug is skipped instead of claimed-then-refused, so it does not trip the capacity breaker or reopen paused/resolved work at teardown. Autopilot…
Release notes on GitHubgald3r v5.0.0-beta.20
Autopilot now runs multiple implementers at once. gald3r autopilot loop / @g-go-go fans out coding agents in parallel by default. Use --no-code-swarm when you need one-at-a-time. Merge-back stays serialized so parallel worktrees do not wedge the git lock.
Release notes on GitHubgald3r v5.0.0-beta.19
The sequential-path survivability release: one loop, honest go-bug, live bucket logs. Silent buckets are killed after 30 minutes by default (was 5); the heartbeat can see bucket output, and a still-running bucket writes a live log. VERDICT: FAIL keeps the bug open. Empty specs fall back to the file/database.…
Release notes on GitHubgald3r v5.0.0-beta.18
Docs match the binary: @g-go-go runs one implementer at a time. Parallel coding is @g-go --swarm, not a hidden loop flag. Field reports (gald3r gfix) warn in the draft itself when GitHub CLI cannot auto-file, fill watchdog evidence from the last captured kill when you type the class by hand, and refuse to auto-file…
Release notes on GitHubgald3r v5.0.0-beta.17
The field-follow-up release: a finished turn is trusted over plumbing noise. A completed no-commit turn counts as real work, so a correct "nothing to change" triage no longer trips the capacity breaker and kills a run that still has queued work. Claims from buckets that die at run-end are released instead of…
Release notes on GitHubgald3r v5.0.0-beta.16
The complete-briefs release: spawned agents now receive their entire instructions, every time, at any size. Agent briefs (coordinator, implementer, reviewer) are delivered whole via file handoff instead of being squeezed through OS command-line limits -- no more truncated instructions on any platform, and every…
Release notes on GitHubgald3r v5.0.0-beta.15
The report-it release: real throughput on every configuration, and a one-command way to tell us when anything still goes wrong. Worker agents are invoked correctly for non-interactive use across every supported agent CLI -- the missing print-mode and trust flags explained both the silent hangs and the "completed but…
Release notes on GitHubgald3r v5.0.0-beta.14
The throughput release: the last field blocker is dead, and projects get a cleaner shape plus three new ways to capture work. Worker agents now receive their FULL instructions (task description, acceptance criteria, working contract) instead of a title-only fragment, and Windows PowerShell-script-installed AI CLIs…
Release notes on GitHubgald3r v5.0.0-beta.13
The restored-playbook release: the autonomous loop got its full operating knowledge back -- and proved it live before shipping. The autopilot coordinator's internal briefing is whole again: restored from the era when multi-hour unattended runs just worked, carried inside the binary, and adapted to the current engine.…
Release notes on GitHubgald3r v5.0.0-beta.9
The throughput release: your subscriptions now do the work. Autonomous implementers on subscription credentials: go-code and go-bug can now drive the same cursor-agent/claude CLI your subscription already pays for -- no API keys required. On a subscription-only machine, autopilot buckets actually implement and fix…
Release notes on GitHubgald3r v5.0.0-beta.12
The eyes-and-teeth release: a stalled swarm can no longer hide, and stopping one actually stops it. A worker that produces nothing gets caught: a per-worker liveness watchdog kills any bucket that goes silent, reports what it captured, and the loop MOVES ON to the next item instead of freezing forever behind one…
Release notes on GitHubgald3r v5.0.0-beta.11
The unblocked release: every kind of repo can loop now. The one-second bucket deaths are over. If your project gitignores its .gald3r/ folder -- the normal setup for application repos -- spawned workers used to land in a workspace with no task specs at all and die instantly. gald3r now detects that shape and points…
Release notes on GitHubgald3r v5.0.0-beta.10
The delivered-orders release: spawned agents finally receive their instructions. The brief actually reaches the agent now. Headless CLI agents (cursor-agent, claude) were being launched without their instructions -- a real agent would run for minutes and change nothing. The prompt is now handed over the way those…
Release notes on GitHubgald3r v5.0.0-beta.8
The observability release: a swarm can never run dark again. Live event stream: every autopilot run now narrates itself the moment things happen -- claims, agent spawns, merges, harvests, breakers, costs -- as [AUTOPILOT][EVENT] lines in your session AND a durable per-run event ledger. New: gald3r autopilot events…
Release notes on GitHubgald3r v5.0.0-beta.7
The harvest release -- and Longship's debut: finished work always comes home, and the Agentic Focus Terminal is now in the box. The harvest gate: work an autopilot agent completes in a worktree is now automatically committed and merged into your checkout -- at every iteration close and once more at run end. A merge…
Release notes on GitHubgald3r v5.0.0-beta.6
The autopilot integrity release: bugs are workable, work is unlosable, parallelism is honest, and local AI works out of the box. New gald3r go-bug: autopilot can now FIX bugs, not just log them -- the full claim, agent turn, self-verify, resolve cycle, wired into deterministic dispatch. A bug backlog becomes…
Release notes on GitHubgald3r v5.0.0-beta.5
Provider routing release: a rate-limited AI provider no longer stops your work. The autopilot coordinator is no longer hardcoded to the Claude CLI: a Cursor host defaults to cursor-agent, and the full routing surface works -- CLI flags (--provider/--model plus per-role --coordinator/--implementer/ --reviewer…
Release notes on GitHubgald3r v5.0.0-beta.4
gald3r install update works again: downloads the per-platform archive, verifies it against SHA256SUMS.txt (fail-closed), extracts, and replaces the running binary safely. Sibling binaries you already have are refreshed too. (This fixes beta.3's known issue -- one last manual install gets you here.) Intel Macs no…
Release notes on GitHubgald3r v5.0.0-beta.3
Autopilot economics release: the swarm now materializes. Deterministic implementer dispatch when a coordinator reports zero implementers despite runnable work (opt out: --no-deterministic-dispatch) Capacity circuit breaker: halts with exit code 3 instead of silently burning budget Empty review queues are skipped for…
Release notes on GitHubgald3r v5.0.0-beta.2
Packaging cleanup: downloads are now one signed archive per platform (plus the .msi/.pkg installers) instead of many separate files. No functional change from beta.1. Download: the latest gald3r_core release (signed, all platforms).
Release notes on GitHubgald3r v5.0.0-beta.1
gald3r is now a fast, signed, native binary on every platform. The engine is rewritten from Python to Go: startup drops from 8-11s to ~120ms, and one signed + notarized binary set ships for Windows, macOS (Apple Silicon + Intel), and Linux - no Python, no runtime to install. Download: the latest gald3r_core release…
Release notes on GitHubgald3r v3.0.0
_Pending release notes accumulate here as tasks and bugs are completed. At publish time this section is renamed to [X.Y.Z] - YYYY-MM-DD and a fresh [Unreleased] block is opened._ Fixed Template mojibake (double-encoded UTF-8) eliminated and guarded (BUG-213). Repaired README.md, CHANGELOG.md, and a…
Release notes on GitHubgald3r v2.4.0
_Pending release notes accumulate here as tasks and bugs are completed. At publish time this section is renamed to [X.Y.Z] - YYYY-MM-DD and a fresh [Unreleased] block is opened._ Added `@g-install-agent` / `@g-install-throne` IDE commands (T1617). Discoverable command wrappers (shipped to every platform that has a…
Release notes on GitHubgald3r v2.3.0
_Pending release notes accumulate here as tasks and bugs are completed. At publish time this section is renamed to [X.Y.Z] - YYYY-MM-DD and a fresh [Unreleased] block is opened._ Headline: the PS1-KILL release. Almost all PowerShell has been removed from the gald3r framework — the hook, skill-script, task-pipeline,…
Release notes on GitHubgald3r v2.2.0
_Pending release notes accumulate here as tasks and bugs are completed. At publish time this section is renamed to [X.Y.Z] - YYYY-MM-DD and a fresh [Unreleased] block is opened._ Added `DeploySystem` — the deploy pipeline step is now Python (T683, PS1-KILL epic T667). Ported the last numbered PowerShell pipeline…
Release notes on GitHubgald3r v2.1.2
gald3r v2.1.2 — Critical upgrade-safety fix If you ever run `gald3r update` / `gald3r upgrade`, update to 2.1.2 now. This release stops the upgrader from touching your data and adds an automatic full backup before every upgrade. Why upgrade now A defect in earlier versions could rename your entire `.gald3r/`…
Release notes on GitHubgald3r v2.1.1
gald3r v2.1.1 A feature release that brings plugins, a richer vault API, and per-platform workflow profiles into the engine, plus identity and hook-wiring cleanups. Highlights Plugins in the engine. gald3r plugin install | remove | list | new | check-compat | update (CLI + MCP tools), backed by a real PluginSystem…
Release notes on GitHubgald3r v2.1.0
# gald3r v2.1.0 — A Real Home, Self-Update & Throne DX Released: 2026-06-20 --- Headline gald3r now has a real home on your machine — one CLI, shared across every project, every OS. Install once. Scaffold projects. Stay updated. Throne tells you when you're out of date and can apply the update from inside the app.…
Release notes on GitHubgald3r v2.0.1
Next Release — gald3r Status: Draft — version number assigned at publish time. Use this file for promotional, enticing copy that will headline the next GitHub release. Technical detail belongs in CHANGELOG.md. User-facing how-to belongs in the README.md pending section. --- Headline _(One punchy line — what makes…
Release notes on GitHubgald3r v2.0.0
The gald3r engine release. gald3r gains a bundled, file-first Python core that backs every system deterministically — while staying 100% markdown-on-disk. Existing installs keep working; the engine is additive, and every slimmed component ships a no-engine fallback. Added Bundled gald3r engine (.gald3r_sys/engine/) —…
Release notes on GitHubgald3r v1.10
Release notes on GitHubgald3r v1.9.0
[1.9.0] - 2026-05-31 (Platforms/ Folder + Post-Push Verify + Plugin System Foundation) Added Plugin system foundation (T1557–T1559): @g-plugin-install lands — install a gald3r plugin from a local path (GitHub-URL path implemented, not yet network-tested). Backed by ADR-015…
Release notes on GitHubgald3r v1.8.0
[1.8.0] - 2026-05-30 (Wiki Launch + GitHub Discussions + Test Harness + 35-Platform Sweep) Added GitHub wiki launched at github.com/wrm3/gald3r/wiki (T1550): 7 pages auto-generated from .gald3r_sys/ — Home, Quickstart, Commands (179 @g-* commands indexed by category), Skills, Agents, Rules, Hooks. Sourced from…
Release notes on GitHubgald3r v1.7.0
Released: 2026-05-28 Version: 1.7.0 Type: Minor -- new features, no breaking changes --- Highlights 34 AI Coding Tools -- Full Sweep 11 new platforms added in this release: Kimi Code (Moonshot AI), TRAE (ByteDance), Qoder (Alibaba), Amp Code (Sourcegraph), Continue, Void Editor, Deep Code (DeepSeek), Kilo Code,…
Release notes on GitHubgald3r v1.6.0 -- WPAC v1.6, Schema System, Ship Skill, Encoding Normalization
Released: 2026-05-25 Version: 1.6.0 Type: Minor -- new features, no breaking changes Previous release: v1.5.0 --- Highlights Schema enforcement system Every .gald3r/ file now carries version provenance, validated against 15 versioned schema definitions. A read-only session-start probe flags drift in under a second,…
Release notes on GitHubgald3r v1.5.2 -- HTML/JSON/TOON output, memory compression, vocab, auto-skill
What's New in v1.5.2 Added HTML output + theme system (g-skl-html-output, --html flag): human-facing reports (g-status, g-review, g-qa-report) now render as self-contained themed HTML. Three built-in themes: gald3r-dark (Catppuccin Mocha, default), gald3r-light (Latte), gald3r-mocha (sepia). Opt-in via output_format:…
Release notes on GitHubgald3r v1.5.0 -- Platform Framework Architecture (.gald3r_sys/)
Released: 2026-05-10 Version: 1.5.0 (Maestro Harvest) Type: Minor -- new features, no breaking changes Previous release: v1.4.0 --- Highlights New security and quality skills `g-skl-security-scan` -- a two-phase SAST scanner. A fast, free regex pass surfaces candidates across 12 vulnerability categories (hardcoded…
Release notes on GitHubgald3r v1.4.0 -- GitHub Copilot, Recon Suite, Release Management
Note: This public repository was created on 2026-05-03. Earlier version releases are documented here for continuity with the internal development history. Added GitHub Copilot support (.copilot/): 6th IDE added to the parity set. Full command surface (89 commands) deployed to .copilot/commands/. gald3r now supports…
Release notes on GitHubgald3r v1.2.1 -- PCAC Spawn, PCAC Send-To
Note: This public repository was created on 2026-05-03. Earlier version releases are documented here for continuity with the internal development history. Added PCAC Spawn skill (g-skl-pcac-spawn, @g-pcac-spawn): spawn a new gald3r project from the current one — creates the project folder in the ecosystem root,…
Release notes on GitHubgald3r v1.2.0 -- Feature Pipeline, Deep Recon, CLI Skills
Note: This public repository was created on 2026-05-03. Earlier version releases are documented here for continuity with the internal development history. Added Feature pipeline (g-skl-features, FEATURES.md, .gald3r/features/): structured staging layer between idea capture and task creation. Features move through…
Release notes on GitHubgald3r v1.1.0 -- Task Circuit Breaker, Status History, Vault
Note: This public repository was created on 2026-05-03. Earlier version releases are documented here for continuity with the internal development history. Added Task circuit breaker ([⚠️] Requires-User-Attention): tasks that fail verification 3 or more times are automatically escalated for human review. Automated…
Release notes on GitHubgald3r v1.0.0 -- Initial Release
Note: This public repository was created on 2026-05-03. Earlier version releases are documented here for continuity with the internal development history. See CHANGELOG.md for details.
Release notes on GitHubgald3r v1.5.1
Added docs/PLUGINS.md: plugin author guide covering skill pack structure, distribution, and naming guidelines for third-party extensions. README "Plugins & Skill Packs" section with naming guideline (plugins are separate works). NOTICE clarified with plugin naming guideline.
Release notes on GitHub
v5.x
v3.x
v2.x
v1.x