Buzz Projects: Git Forge on Your Own Nostr Relay

Key takeaways

1. Date / source. August 18, 2026, Block Engineering blog, ~1,800 words, signed by Thomas Petersen (Principal Designer & Builder). No outbound links, no figures.
2. Key framing. Buzz Projects is a forge in which conversation is a first-class artifact — projects are linked to channels, and « Instead of reconstructing why something happened after the fact, the history is already there. » ### The trust model: two paragraphs meant to be read together | Post section | Verbatim | What it establishes | |---|---|---| | First Class Citizens | « No forced guardrails, no limitations on what your agents are allowed to help you with. No limits to how much you can delegate to your agents in the network. » | removal of prior constraint | | Weaving it all together | « Every push, review, approval, and merge is a signed Nostr event. If an agent authors a patch, you can see which agent produced it and which human authorized that agent to act. » | signed record of the act | | same paragraph, closing | « we are already exploring ideas around agent trust protocols informed by past behavior » | deriving trust from history | The position is architectural: in a network where delegation is unlimited, a priori authorization scales poorly, signing does not. It is the "network" version of what [[longwell-block-buzz-workspace-agents-nostr-2026-07-21]] laid out in July at the identity level (« authorization does not erase authorship »). Two limitations the post does not address: an ex post record does not prevent the first instance of harm, and a signed history proves what was done, not the completeness of what is shown — a negative register (rejected PRs, regressions, incidents) would be necessary for a reputation, and is not mentioned. ### What the signature guarantees, and what it doesn't | Guaranteed | Not guaranteed | |---|---| | the event was indeed produced by this key | that the key produced everything shown of it | | the integrity of each isolated event | the completeness of the batch presented | | the authorizing agent → human link | the discoverability of events outside their origin relay | Architectural point not to be attributed to this post: the relay is the single source of truth, with no replication or peer-to-peer exchange. This comes from Block's ARCHITECTURE.md as consolidated in [[buzz-block-panorama-deep-research-2026-08-12]], not from this text, which says nothing about it. Consequence: the sovereignty on offer is an exit right, not an availability guarantee. ### Git interoperability: the most verifiable claim Verbatim: « These are standard git repositories, like you already know them… You can fetch, clone, pull, and push over plain Smart HTTP, with no custom tooling or wrapper CLI required. Your Nostr key is your identity throughout the process… You do not need another account, another identity, a separate token, or a GitHub account connected in the background. » Consequences: exit cost is low by construction (a git clone retrieves the code), and the single identity removes token management. Nuance to retain: only the code is standard — conversation, review, PR↔channel linking, and contribution history live in Buzz-specific Nostr event kinds. A test doable in ten minutes: clone then push from a bare Git client against a Buzz relay, without the buzz CLI. ### Announced inventory vs. delivered inventory The opening paragraph lists six fragments to be brought together; the « What's next? » section lists what exists. | Fragment named in the opening | Present in the August 18 inventory | |---|---| | « A bug report lands in one tool » | yes — issues | | « The discussion happens in another » | yes — channels linked to the project | | « The fix lives on a branch somewhere else » | yes — hosted git, multi-repo | | « Review happens in a comment thread attached to a diff » | yes — PRs, reviews, inline comments | | « CI runs in another system » | no — cited among the promises, absent from the inventory | | « Release notes get written later » | no | Six fragments announced, four delivered. Do not reuse the list "repos, branches, pull requests, issues, CI, or all the related conversations" as an inventory of shipped features. ### Distinction worth reusing: executing is not the same as being present | What a terminal gives an agent | What network presence adds | |---|---| | execute, write files | a stable identity others can address | | local, disposable state | an attached history that survives the session | | act | being held accountable for what was done | A useful frame for weighing a tool-agent (invoked) against a member-agent (addressed). Corollary the post does not address: a persistent presence is also a persistent attack surface — see [[valente-zalewski-beyond-zero-enterprise-security-ai-era-2026-07-20]]. ### "Context is (almost) all you need" Verbatim: « The more context we can make available, the better equipped the agents are, the more intelligent the whole network becomes. » Monotonicity is asserted without argument: neither the cost of context, nor its degradation, nor its selection is discussed, and the « (almost) » is never spelled out. Worth comparing to the token budget quantified at Block itself by [[patel-block-buzz-teams-tokens-benchmarks-2026-08-06]]. ### The post's only claim of an effect « We've already seen a number of examples where agents understood context humans didn't and helped course correct otherwise unproductive directions. » No example, no number, no success criterion. Correct phrasing for reuse: « Block reports, without documenting them, cases where agents redirected unproductive directions using channel context. » Fiche to reopen if Block publishes these cases. ### Citation hygiene 1. Citable as Block's architectural intent; not as proof of functioning or of adoption. 2. Always carry the stated maturity: Experiments tab, « fairly elementary », Buzz « still in beta ». 3. Distinguish features shown in screenshots, those inventoried under « What's next? », and those only named in promise sentences (CI, release notes). 4. The « agent trust protocols » are a stated direction, with no schema, no criterion, no timeline.