What shipped — Jun 28–Jul 5, 2026

Update · August 10, 2026: the tools in this post have new names. warden is now redstamp, canon is now truecopy, keeper is now strongroom, and picket is now fieldpass. Same tools, same code. This post keeps the names it shipped with; the repo links redirect to the renamed homes.

dario: 30 PRs, 33 releases (v4.8.102 → v4.8.134)

Claude Sonnet 5 support landed the day it was available. Fable 5 support is back, the model returned globally on Jul 1 after its June suspension was lifted, and dario dropped its defensive model filter the same week. A new headless admin API shipped for scripted account management. The rest was reliability: SDK-drift absorption via dario's own self-healing release pipeline, 33 releases in seven days. How the pipeline heals itself →

warden: 25 PRs, release 0.3.0

Four new framework adapters this week: CrewAI, LangGraph, the OpenAI Agents SDK, and Microsoft AutoGen, all governed at the MCP layer, none of them patched to do it. warden also shipped Arena, a neutral firewall benchmark mapped to MITRE ATT&CK, instead of a leaderboard where we conveniently win. Warden's own number on it: 96% recall, 100% precision, deterministic. False-positive fixes were validated against 2,000+ live marketplace audits. One gate, four frameworks → · The leaderboard I refused to build →

canon: 10 PRs, plus canon-action@v1

Five feature PRs, and the reason canon exists: a 2,019-skill audit of the official Claude Code marketplace plus nine community ones, then a follow-up covering all 66,541 skills in ClawHub, the open registry where the poisoned-skills incident that started this category actually happened. Zero confirmed malicious in either. The harder part was getting the scanner's false-alarm rate from ~10% down to 0.6%, measured against the corpus, not a demo. New this week: canon-action@v1, the same scan as a composite GitHub Action you can drop into CI. The marketplace audit → · The ClawHub scan →

keeper: 15 PRs

Master-key rotation shipped. The bigger story is the migration case study: our own fleet went from 132 inherited environment keys down to 13 kept and 119 replaced with scoped, revocable leases, git tokens and the model key included. Found and fixed a reflected-secret bug in keeper along the way, proven against a live echo endpoint. The migration case study →

hands: 10 PRs, v0.19.0 + v0.20.0

Two releases for hands: hands daemon (persistent, unattended automations) and auto-crystallize, where tasks you repeat promote themselves into $0 macros without being asked.

picket: 9 PRs, v0.2.0

picket v0.2.0 shipped Jul 2. The week went to property-based fuzzing of the injection-detector plane and a security baseline: CodeQL, Scorecard, Dependabot, least-privilege CI tokens. The picket launch post →

cordon: 7 PRs

All internal this week, redaction-path hardening and fuzzing in cordon, no user-facing surface changes. Worth saying plainly instead of finding something to spin.

Also shipped

deepdive v0.27.1 + v0.28.0 (relevance-windowed per-source caps, 16 PRs) · hybrid v1.6.0 + v1.6.1 (batched self-consistency voting, answer cache, 17 PRs) · browser-bridge v0.2.0 (built-in token-auth CDP proxy, 3 PRs) · amnesia (search perf + engine-recovery fix, 2 PRs) · brand and stack refreshes across the org profile (8 PRs).

Numbers are public-repo only, pulled directly from GitHub. The platform repo that runs askalf's own operations is private, so it's excluded on purpose, not by oversight.

Install / upgrade

npm i -g @askalf/dario@latest     # standalone dario
npm i -g github:askalf/canon      # security tools install from GitHub

The security tools (warden, canon, keeper, cordon, picket) currently install from GitHub, see each repo's README for its one-liner. Registry packages land once the npm publish issue clears.

askalf is self-hosted. When something breaks, we say so here.

Early-access waitlist → askalf.org. Public claims above are verifiable on GitHub and npm (as of 2026-07-05).