📦 Shipped
Weekly recap of public-repo activity across prime-radiant-inc and obra.
-
evener cuts three releases on an LLM-gateway and backlog-issue sweep, while superpowers retunes brainstorming for v7.0.1
Partial week — data runs through Friday afternoon, not the full calendar week. prime-radiant-inc/evener tags three releases (v0.3.14, v0.3.15, v0.3.16) across a sustained sweep of LLM-gateway compatibility work, a provider-connection UI, and a long tail of numbered-backlog-issue test hardening. obra/superpowers tags v7.0.1, retuning its brainstorming skill after community feedback on v7.0.0 and restructuring three long skills to survive Claude Code's compaction cutoff.
-
evener tags v0.3.8 after a week-long copy-oracle test-hardening push, and evener-client-playdate debuts as a native Playdate handheld client
Partial week — data runs through Friday evening, not the full calendar week. prime-radiant-inc/evener tags v0.3.8 after a week running a sustained copy-oracle test-hardening campaign, a web-UI widget-library migration, device-pairing work, and a hardened per-test timing gate. A new repo, prime-radiant-inc/evener-client-playdate, also debuts as a native Playdate handheld client for evener with voice composition and direct Tailscale connectivity.
-
evener tags v0.3.7 and finishes a generated memory index, superpowers-chrome defaults to redacting secrets exposed at page load, and a new repo debuts a morning-routine board
Partial week — data runs through Friday evening, not the full calendar week. prime-radiant-inc/evener tags v0.3.7 and finishes a generated memory index, with two of three notification-semantics cleanup PRs landing; obra/superpowers-chrome ships v3.0.11, defaulting to redacting secrets present at page load; and a new repo, obra/morning-routine-board, debuts as a from-scratch chore-checklist PWA built in a single day.
-
evener ships v0.2.0 through v0.3.4, superpowers-chrome closes a credential-marker leak in eval/extract/attr, and agentic-usage-meter adds banked-reset visibility
This is a partial week — data runs through Friday evening, not the full calendar week — in which prime-radiant-inc/evener, prime-radiant-inc/agentic-usage-meter, and obra/superpowers-chrome shipped tagged releases, and obra/muse-skills launched as a new repo of automated daily skill snapshots.
-
blogosphere cuts its first release behind a signed macOS build pipeline, superpowers and superpowers-chrome each ship a point release, and evener picks up an outside contributor's code-signing work
The most recent week, still in progress: obra/blogosphere cuts its first-ever tagged release, v0.1.0, behind a from-scratch signed-and-notarized macOS release pipeline; obra/superpowers ships v6.4.2 and obra/superpowers-chrome ships v3.0.6. prime-radiant-inc/evener has its busiest week by commit count, and picks up a handful of outside-contributed commits doing macOS code-signing CI work. Partial week — data runs through Friday evening, not the full calendar week.
-
teststrip ships v0.3.0 and v0.3.1, agentic-usage-meter cuts v0.2.6, and clearance pulls in four outside contributors in one week
The most recent week, still in progress: teststrip ships two point releases behind a heavy scenario-testing push, agentic-usage-meter cuts v0.2.6 with a MiniMax quota fix, and evener has its busiest week of the month. prime-radiant-inc/clearance has an unusually open week, landing four separate outside-contributor PRs, and three more outside contributions turn up elsewhere — on awesome-superpowers (Matías Beccaria), prime-radiant-marketplace (caioniehues), and obra/superpowers (GoldJohnKing). Partial week — data runs through Friday evening, not the full calendar week.
-
study-skills, shepherd-pr, and scan-to-model all launch, and episodic-memory ships v1.5.0 and v1.6.0 back to back
Three brand-new repos land in one week — prime-radiant-inc's study-skills (durable agent learning) and shepherd-pr (PR shepherding), and obra's scan-to-model (scan-to-Blender reconstruction) — while obra/episodic-memory cuts two releases in the same day, adding a fifth conversation harness and an auto-sync off switch. evener is the busiest repo again, and picks up its first outside contributors in a while.
-
A light week: evener's busiest repo again, superpowers-evals' campaign push
A light week: no new repos, no tagged releases. evener is the busiest repo again, working through a provider-registry migration and a long tail of hub/agent/web hardening; superpowers-evals runs a fast-moving campaign-infrastructure arc. alltheagents.org gets a polish-and-launch-prep pass, and gauntlet and lace each land a single small fix. Partial week — data runs through Friday evening, not the full calendar week.
-
automatic-sshfs's first real release, three winpepper alphas, and evener's REST-to-AppWire migration
The most recent week, still in progress: automatic-sshfs cuts its first real release (known_hosts enumeration, reliable polling), winpepper ships three alpha releases fixing a dead-end first-run install and trimming its default footprint, evener spends the week migrating REST endpoints to typed AppWire, and a new toy AArch64 C compiler shows up on obra with an open question about who — or what — is driving it.
-
Two renames and a new mount: evener (formerly serf), ledger's move to chit, and automatic-sshfs
The most recent week, still in progress: automatic-sshfs goes public and gets built partly by its own coding-agent tooling, ledger cuts v0.2.0 and v0.3.0 while renaming itself to chit at the packaging layer, winpepper cuts an outside-contributed alpha, and serf completes a full rename to evener across a very large week of commits.
-
Tools that help other tools ship: everyharness, proving-it-works, and superpowers v6.3.0
The most recent week, still in progress: three new repos go public (everyharness, everyharness-container, proving-it-works), superpowers cuts v6.3.0 with Devin and Hermes support, clipfan and agentic-usage-meter both release, and serf runs a top-to-bottom UX pass — with outside contributions from recognizable names.
-
The widest week: releases across the board, and the whole org gets mapped
The widest week in the window by repo count: obol, agentic-usage-meter, and superpowers-chrome all cut releases, awesome-superpowers opens, serf logs a third four-figure week — and an org-wide service-catalog sweep touched nearly every repo at once. We count the sweep honestly and don't let it masquerade as product work.
-
Building for a community: a code of conduct and a usage meter
Prime Radiant publishes a community code of conduct and opens agentic-usage-meter, a macOS quota tracker. No releases this week — but two new repos that say where things are heading, plus another heavy serf week and a research campaign that closed on a substantial cost cut.
-
superpowers v6.2.0, winpepper goes streaming, and serf's biggest week
superpowers cuts v6.2.0, winpepper ships two alpha releases and adds cloud ASR, and serf logs its heaviest push of the window on session and pane architecture. Repos active across both orgs, with outside contributors landing on both superpowers and winpepper.
-
Two tools from scratch: blogosphere and smevals
obra/blogosphere — a local-first blogging client — and smevals, a small-model eval framework, are both built from nothing in a single week. teststrip and obol cut releases, and the contributor list widens across both orgs.
-
First releases: books-for-bots, obol, and teststrip go public
Two Rust tools built for AI agents — books-for-bots and obol — cut their first tagged releases, teststrip's macOS photo-culling app goes public with releases of its own, and under the surface serf runs a fuzz-coverage campaign and smevals gains a contribution from Simon Willison.
-
Two new repos go public, and superpowers turns to Codex
A Linear-tracking plugin and a native-Linux terminal workspace manager both go public, superpowers ships two Codex-focused releases, and under the headlines serf runs a coverage-hardening campaign while teststrip races through its culling workflow. Active repos across both orgs.
-
serf goes public, toil arrives
serf ships v0.1.0, toil lands as a new workflow orchestrator, and a credential-axis push reshapes the eval lab. Active repos across both orgs.
-
Superpowers 6 ships, claude-session-driver hits 4.0
A release-train week: Superpowers 6 lands, claude-session-driver ships its TypeScript rewrite as v4.0.0, and obol and slackline both cut new releases. The last week before our first published recap.
-
clipfan reaches 1.0, and a research-methodology stack takes shape
clipfan crosses v1.0.0, three new repos appear — a plugin-stats scraper and the superpowers-autoresearch / superpowers-docs evidence-methodology pair — and a full Quorum TypeScript port lands in the eval lab.
-
obol arrives, clipfan grows a self-healing mesh
obol launches — a multi-language agent-cost estimator — homedir-manager splits out of dotfiles, and slackline, clearance, and clipfan all cut releases. A launch-and-release-heavy week across both orgs.
-
clipfan launches, winpepper finds its voice
clipfan ships its first tagged release — a Mac-to-fleet clipboard daemon — and winpepper's local dictation reaches v0.6.1-alpha. A broad week across both orgs, from eval-lab economics to VM-orchestrator surgery.
-
slackline and episodic-memory cut releases
episodic-memory hardens its cross-platform install, slackline ships default-recipient fixes, and a broad week spans a serf TUI redesign, a Codex OAuth provider in sprout, and a third VM backend in stockyard — the last stop before the blog's live coverage begins.
-
winpepper is born, claude-session-driver hits v2
winpepper's Windows dictation tool is created and built out in a single week, claude-session-driver drops its fake approval gate for a real v2, episodic-memory lands end-to-end Codex support, and superpowers-evals splits out as its own lab.
-
bot-toolkit and parallel-adversarial-review launch
Two new repos ship to npm and GitHub the same week they're created, superpowers cuts v5.1.0 and seeds its eval harness, and streamlinear cuts its first packaged release.
-
books-for-bots arrives, episodic-memory hotfixes
A new Rust CLI turns EPUBs into agent-navigable Markdown, episodic-memory ships three point releases ending in an emergency hotfix, and outside contributors show up early across both orgs.
-
sprout builds a memory, ts-libghostty binds a terminal — and watches NetHack
The batch's biggest week: sprout stands up 'mira,' a full agent-memory system, across ten phases; ts-libghostty goes from a spec to a released TypeScript binding stack over Ghostty's terminal engine, topped by an agent that watches an LLM play NetHack; gauntlet grows an evidence-and-transcript system; and several tools get flattened and licensed for a new public marketplace.
-
gauntlet reaches v1.5, hearthstone grows an admin and locks down
gauntlet lands its v1.5 transition — server-owned runs, a real configuration system, and browser isolation — while hearthstone matures on the operations side with an admin console, careful house deletion, rate limiting, a security pass, and a relicense. Plus a new Kindle-highlight exporter and tooling to mirror superpowers as a Codex plugin.
-
hearthstone learns Drive and PIN sign-in, a methodology proves itself
hearthstone keeps its blistering pace — Google Drive import, PIN-and-QR sign-in, multi-household support, and a first cloud deploy — while iterative-development grows from a scaffold into a real spec-driven methodology and a companion repo dogfoods it by building a working macOS dictation app across twenty-one iterations.
-
hearthstone builds a household RAG, mutter-butter stamps your windows
A ninety-commit build stands up an entire household document-Q&A product — backend, iOS app, and an evaluation harness — while a new GNOME extension turns minimized windows into draggable 'stamps,' and stockyard tunes a macOS microVM down to a 1.24-second cold boot.
-
serf tunes its way to 82% on terminal-bench, pepper-x brings voice dictation to GNOME
serf spends 126 commits tuning its agents against terminal-bench 2.0 — from ~54% to a final 82% — a new leaderboard-analysis tool grows up alongside it, and pepper-x arrives as a GNOME voice-dictation app.
-
serf learns task dependencies, sprout tunes for mixed models, slackline and knowledge-graph arrive
serf gains a dependency-aware task system, sprout spends the week hardening delegation across mixed models, and two new projects land: slackline, a Go CLI for Slack, and knowledge-graph, a local vault indexer with an MCP server.
-
superpowers reaches v5.0, sprout adds live pricing, stockyard gets command queues
superpowers ships a major v5.0 release with visual brainstorming and a zero-dependency server, drawing contributions from half a dozen people outside the org; sprout wires in live model pricing; and stockyard builds a custom Firecracker kernel and a command-queue system.
-
clearance ships a v1 editor, sprout restructures its genome, and agent-eval tooling takes off
clearance — a native macOS markdown editor — goes from bootstrap to v1.1.0 in a week, sprout restructures how its genome resolves agents, and a cluster of agent-evaluation tools (gauntlet, a new harbor dashboard, serf's eval instrumentation) all advance together.
-
sprout's 545-commit sprint, serf grows a TUI, and a wave of new agent tooling
sprout — a self-improving agent framework — lands the single largest repo-week in the whole backfill at 545 commits, serf gains a terminal UI and model picker, the engineering-notebook gets a three-panel redesign, and a cluster of new plugins and tiny tools ship alongside external contributions.
-
serf grows a TUI, sprout begins, and the engineering notebook lands
serf's biggest week yet — a full terminal UI, append-only transcript logging, and live model listing across 120 commits. A second agent framework, sprout, starts from scratch with a self-rewriting genome, and a new engineering-notebook turns session logs into a searchable journal.
-
serf begins, Wayback recovery, Beeper sync, llm-proxy adds Bedrock
An agent framework called serf lands its foundation in a single enormous week, a Wayback Machine recovery tool and a Beeper message archiver both ship, llm-proxy gains AWS Bedrock support, and superpowers reaches v4.3.0 with outside contributors.
-
The session viewer arrives, superpowers v4.2.0, scenarios tests prompt injection
A session-explorer web app lands in a single week of heavy building, superpowers ships v4.2.0 with Windows fixes and Codex native skill discovery, and the scenarios project turns into a real prompt-injection test suite.
-
Moltipass takes shape, Scribble migrates to the Agent SDK, llm-proxy's release train
A new iOS client for Moltbook lands its first build, the Scribble bot migrates onto the Agent SDK with a dual-session architecture, and llm-proxy shipped five releases in a week as it grew agent-observability features.
-
Scribble comes online, faster VM boots, llm-proxy v0.2.0
A Slack-native assistant bot lands its first orchestrator, the Firecracker VM stack roughly halves its boot time, llm-proxy ships Loki export in v0.2.0, and superpowers reaches v4.1.x with outside contributors turning up across the ecosystem.
-
stockyard's foundation lands, lace gains skills and todos, llm-proxy v0.1.0
stockyard's microVM-orchestration daemon gets built from scratch in a 329-commit week, lace grows a skills system and task-tracking tools, and llm-proxy — a Go logging proxy for LLM APIs — cuts its first release.
-
lace becomes standalone: core dissolution, async jobs, ACP alignment
lace's biggest week yet — the agent package absorbs core and stands alone, the protocol aligns to ACP, an async background-jobs system lands, and the TUI gets a ground-up redesign.
-
lace takes shape; Winby ships to 1.3
The agent runtime lace is built up from its protocol in a single week, and Winby — a macOS window manager — goes from initial commit to v1.3.1 across eight releases.
-
streamlinear: the first Prime Radiant repo
The company's public record starts here — prime-radiant-inc's first repository, streamlinear, a Linear MCP tool — while the obra side cuts two superpowers releases and outside contributors turn up across the ecosystem.