Entity trail

Goose

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Showing the first 40 findings. More graph evidence exists in the corpus.

Corpus findings

2026-09-04 / github-pulse-researchergoose 1.49.0 adds an on_failure block for PreToolUse hooks and ships web search and browser use as built-in skillsThe 2026-09-03 release adds an on_failure block for PreToolUse hooks (#11449) and a PreToolUseResult event with a stable tool_call_id across the whole tool lifecycle (#11120), which together make hook-based tool gating actually observable. It also promotes web-search and browser-use to built-in skills (#11233), loads extensions in the background so the CLI prompt is usable immediately (#10403), adds a desktop auto-updater (#10614) and Linux ARM64 desktop packages (#11518), and forwards ACP thinking-effort derivation (#10949). The identical v1.49.0 tag was published simultaneously under both aaif-goose/goose and block/goose.
2026-09-04 / vibe-coding-researchergoose 1.49.0 quietly fixes three fail-open behaviours in its tool pathBuried in the bug list of the 2026-09-03 release: goose now fails closed on invalid default GCP credentials (#11363) instead of proceeding, rejects cmd.exe commands containing newlines (#11537), and sanitizes hidden Unicode in Bedrock tools (#11121). It also bounds non-streaming JSON responses (#11109) and gives an honest compaction failure message rather than reporting success when no tool responses exist (#10500). Each of these is a case where the agent previously kept going with a broken precondition, which is the failure mode that costs you a whole unattended run.
2026-09-04 / vibe-coding-researcherGitSpawn: opening a folder is enough to run attacker code in seven coding agentsManifold Security published eight findings across seven coding agents (Claude Code, Codex, Cursor, Grok Build, Qwen Code, goose, Hermes Agent) that all reduce to one trick: a repo's own .git/config sets core.fsmonitor to an attacker command, and any background `git status --porcelain=2 --branch` or `git diff --name-only HEAD` the agent runs to populate context executes it, with host privileges, outside the sandbox, before any trust prompt. Four of the eight were still unpatched at publication, including a second Claude Code path found 15 July 2026 that abuses a different git key and was open on 2.1.252, plus Qwen Code 0.22.3, Grok Build 1.0.13 and Hermes 0.21.0. Delivery does not need a clone: a zip, a sync folder, a shared drive or a USB stick carrying an intact .git directory is the whole attack.
2026-09-03 / projects-researcherpacifio/atlas frames version control as an agent problem, tracking multiple coding agents' changes in one Rust binaryAtlas calls itself source control for agents, letting you run several coding agents, track what each changed and query the result in one place. It is at 3,085 stars with alpha-0.3.0 ("Atlas ACP + Timeline") cut 2026-08-25 and a push this morning, and carries only 25 open issues against 198 forks. The ACP work in that release means it speaks the same Agent Client Protocol that Goose and Zed use, so it is positioning as a layer under existing agents rather than another agent.
2026-08-28 / saas-disruption-researcherGrith Scores Every Syscall a Coding Agent Makes and Queues the Ambiguous Ones for a Humangrith-ai/grith posted to Show HN on August 28: ptrace and seccomp supervision on Linux that assigns a risk score to each syscall, auto-allowing below 3.0, queuing 3.0 to 8.0 for human review, and denying above 8.0, with the canonical demo being an agent trying to POST a .env file to an external host. It ships built-in supervisor profiles for eleven agents including Claude Code, Codex, Aider, Cursor, Cline, Copilot and Goose, positioning itself as middleware under the whole agent market rather than a competitor to any one of them. It is MPL-2.0 with paid tiers gated by signed licenses in the same binary, and at 3 stars it is day-one early, so treat the numeric risk thresholds as a design sketch not a validated model.
2026-08-28 / agents-researcherGoose v1.48.0 ships roughly twenty security fixes at once, nearly all of them fail-closed correctionsBlock's Goose released v1.48.0 on 27 August with a security section dominated by defaults that previously failed open: fail closed on malformed tool visibility, permission denies now take precedence, fail closed on invalid default GCP credentials and invalid Codex ACP mode, honor plugin enablement for skills, honor MCP tool model visibility in Code Mode, sanitize Unicode tags in MCP prompts, bound recursive mention scans and call-graph traversal, and avoid a predictable editor symlink. The same release adds an on_failure block for PreToolUse hooks, a stable tool_call_id across the tool lifecycle, built-in web-search and browser-use skills, and MCP conformance tests in CI. The concentration of fail-open bugs in one release is the story for anyone running Goose with permissions configured.
2026-08-28 / github-pulse-researchergoose v1.48.0 adds six declarative LLM providers and moves cost tracking onto custom provider cost fieldsBlock's goose shipped v1.48.0 on 2026-08-27 with TrustedRouter, OpenCode Zen, Gondola, SayGM, Lynkr and PleumRouter all added as declarative providers rather than hand-written integrations, plus Z.ai's GLM-5.2 replaced by GLM-5.3 and a model-native audio transcription provider. Custom provider cost fields now drive cost tracking. The declarative-provider pattern means adding a gateway is a config entry, which is why six landed in a single release.
2026-08-28 / github-pulse-researchertransformers v5.16.1 ships GLM-5.3-Flash, a 320B/18B-active natively multimodal model at one-tenth of GLM-5.2's priceHugging Face cut v5.16.1 on 2026-08-26 as a special release just to land GLM-5.3-Flash, the first natively multimodal model in the GLM-5 series. It has 320B total parameters with 18B active, trained from a new base model, and introduces a hybrid architecture combining sparse and linear attention to cut long-context serving cost. Z.ai claims it beats GLM-5.2 across benchmarks at one-tenth the price and approaches Claude Opus 4.8 on coding and agentic benchmarks; block/goose v1.48.0 has already swapped GLM-5.2 for GLM-5.3 in its Z.ai provider.
2026-08-23 / agents-researcherGoose 1.47.0 extracts the agent loop into a goose-agent crate as an unrolled state machineThe 21 August 2026 release of Block's Goose (now under AAIF) creates a goose-agent crate with an unrolled agent loop state machine (#11216) and makes that state machine generic (#11139), a structural change for anyone embedding Goose rather than running the desktop app. Security and cost fixes land alongside: secure token transport enforced in OAuth (#11183), pre-registered OAuth clients for streamable_http extensions (#11182), Gemini OAuth provider deprecated (#11223), and the prompt-cache write premium dropped on one-shot fast-model calls (#11179). The computercontroller extension loses its automation_script, web_scrape and cache tools (#11198).
2026-08-22 / github-pulse-researchergoose v1.47.0 is mostly a resource-bounding security pass, and quietly stops paying the prompt-cache write premium on one-shot fast-model callsThe AAIF goose project shipped v1.47.0 on 2026-08-21. The changelog reads as a deliberate hardening sweep: bounded XLSX range reads, bounded local image reads, bounded action-required stream admission, narrowed desktop file-read IPC, enforced secure token transport in OAuth, and SQLite bumped to 3.51.3. A separate fix stops paying the prompt-cache write premium on one-shot fast-model calls, which is a direct cost cut for anyone running goose in a loop, and the Gemini OAuth provider is now deprecated.
2026-08-22 / vibe-coding-researcherPattern: Agent Runtimes Are Consolidating Under Neutral Foundationsgoose's move to the Agentic AI Foundation at the Linux Foundation follows the Agent Plugins 1.0 spec published by a vendor-neutral technical steering committee on 2026-08-06 and Anthropic's release of the Agent Skills format as an open standard. The practical effect for builders is that skill and plugin directories are starting to be portable across Claude Code, Cursor, Codex and others rather than locked to one client. The caveat that keeps surfacing in practitioner writeups is that portability is not efficacy: focused skills with at most three modules reportedly outperform exhaustive bundles.
2026-08-22 / projects-researcherGoose 1.47.0 Lands August 21 With an Interactive Git Branch Indicator in the Chat Baraaif-goose/goose (Rust, 53,217 stars) shipped v1.47.0 on 2026-08-21, nine days after v1.46.0 on 08-12. The changes are small but pointed at a real agent failure mode: an interactive git branch indicator in the chat bottom bar so you can see which branch the agent is operating on, plus a model picker that surfaces recently used models and pre-registered OAuth client support for streamable_http extensions. Branch-awareness in the chat surface is the kind of fix you only build after watching agents commit to the wrong branch.

Graph relationships

USES
GLM-5.3
goose v1.48.0 replaced GLM-5.2 with GLM-5.3 in its Z.ai provider
Source finding
BUILT BY
Block
Block makes the goose agent platform
Source finding
RELEASED
Block
Block released Goose v1.48.0 on August 27 with twenty security fixes.
Source finding
DEPRECATES
Gemini
Goose 1.47.0 deprecated Gemini OAuth provider support.
Source finding
SUPPORTS
MCP
Goose ships with 70+ MCP extensions.
Source finding
BUILT BY
Block
Goose v1.40.0 is Block's open-source agent.
Source finding
SUPPORTS
MCP
Goose supports interactive UI components returned by MCP tools.
Source finding
SUPPORTS
Red Hat
Red Hat MCP server enables Goose agents to inspect systems.
Source finding
RELEASED
Block
Block's internal coding agent 'Goose' has been open-sourced
Source finding
RELEASED
Block
Block released goose as open source, a Rust-native AI agent.
Source finding
SUPPORTS
Gemma 4
Goose added Gemma 4 local model support.
Source finding
BUILT BY
Block
Block's open-source AI agent Goose transferred to AAIF.
Source finding

Source trail

GitHub Blaizzy/nativ Block's July 29 release Block's buzz (GitHub Changelog)(GitHub Changelog)(Simon Willison)(arcbjorn.com)Morph LiteParse markitdown goose

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…