Agent Pulse / Daily decision brief / 2026-07-29

Treat a shared agent transcript as a deployment boundary

Qwen Code's fresh session-writer work makes one bounded, project-scoped call: if an ACP or daemon deployment lets more than one process resume the same persisted agent session, test the ownership and recovery boundary before enabling any concurrency feature. The project now documents an experimental writer lease that is default-off, fixed at process start, and unsafe across mixed settings or versions; its adjacent recovery work shows why the boundary matters after shutdown.

One decision-worthy insight. Action, effort, and rollback. 24 / 24 included projects monitored. Methodology ›

Today's Action

Test / Project-specific claim
action

Treat a shared agent transcript as a deployment boundary

What changed. The inspected opt-in adds `experimental.sessionWriterLease` for ACP and daemon recorders, pins its value when the process starts so a settings reload cannot mix leased and legacy writers, and documents cross-process fencing. The related shutdown change adds writer-lock release and recovery work, rather than treating a process exit as proof that the next agent can safely resume the same history.

Why it matters. A shared transcript is more than a chat log when an agent can resume, branch, run tools, or make a follow-up decision from it. Two writers that disagree about ownership can turn a recovery path into a silent history fork. A named deployment boundary makes that risk testable instead of leaving it as an accidental property of timing.

Operator of Qwen Code ACP or daemon sessions with persisted histories 30–60 minutes

Test one controlled writer hand-off

In an isolated test workspace, start one session owner, force a second ACP or daemon process to request the same persisted session, then close or interrupt the first owner. Verify that the second writer is refused while ownership is live, can resume only after the first has released or recovered, and that a mixed setting or old process is drained before any opt-in rollout.

Reversible

Evidence Chain

2 direct receipts
Confidence / strong

Two fresh, directly inspected Qwen Code commits describe the same session-ownership boundary: one supplies a constrained rollout switch and the other supplies its shutdown and recovery mechanics.

Limit

These are source-level Qwen Code receipts. They do not establish a release containing the feature, interoperability with every ACP client or filesystem, or that an operator's own transcript storage and backup process are safe.

What would change this

Revise this call if a later public test or issue shows two opted-in writers can concurrently modify one persisted session, if a shutdown leaves an unrecoverable lock, or if the documented setting semantics no longer require a process restart and coordinated rollout.

Watch next. Watch for a public end-to-end two-process recovery test, release notes that declare the lease ready beyond experimental use, and evidence that mixed-version deployments receive a safer migration path.

Secondary Signals

Context, not recommendations

These source movements did not become today's decision brief. They remain visible so readers can inspect the receipts without mistaking activity for consequence.

QwenLM/qwen-codeOpen receipt ›

QwenLM/qwen-code

QwenLM/qwen-code shows 284 public commits this week, down 32 from the prior week and 5.4 above its current 30-day pace, with tools as the strongest subject signal.

NousResearch/hermes-agentOpen receipt ›

NousResearch/hermes-agent

NousResearch/hermes-agent shows 2,018 public commits this week, up 678 from the prior week and 738.2 above its current 30-day pace, with ux as the strongest subject signal.

Kilo-Org/kilocodeOpen receipt ›

Kilo-Org/kilocode

Kilo-Org/kilocode shows 181 public commits this week, down 61 from the prior week and 117 below its current 30-day pace, with tools as the strongest subject signal.

Tracking Health

24 / 24 monitored today

24 freshly tracked / 13 of 13 core coding agents / 3 excluded with reasons / 0 unavailable / 0 stale. All 27 watchlist repos have an explicit decision.

Read all 27 tracking decisions
openai/codex Tracked

Included in Agent Pulse with a readable, recently refreshed public-git source.

Jul 29, 04:30 AM GMT+2

Included in Agent Pulse with a readable, recently refreshed public-git source.

Jul 29, 04:30 AM GMT+2

Included in Agent Pulse with a readable, recently refreshed public-git source.

Jul 29, 04:30 AM GMT+2

Included in Agent Pulse with a readable, recently refreshed public-git source.

Jul 29, 04:30 AM GMT+2

Included in Agent Pulse with a readable, recently refreshed public-git source.

Jul 29, 04:30 AM GMT+2

Included in Agent Pulse with a readable, recently refreshed public-git source.

Jul 29, 04:30 AM GMT+2

Included in Agent Pulse with a readable, recently refreshed public-git source.

Jul 29, 04:30 AM GMT+2

Included in Agent Pulse with a readable, recently refreshed public-git source.

Jul 29, 04:31 AM GMT+2

Included in Agent Pulse with a readable, recently refreshed public-git source.

Jul 29, 04:33 AM GMT+2

Included in Agent Pulse with a readable, recently refreshed public-git source.

Jul 29, 04:33 AM GMT+2
git/git Excluded, with reason

Watched for source-control infrastructure stories, but excluded from Agent Pulse project rows because it is a substrate rather than an AI-agent project and its general development volume would distort the editorial filter.

Editorial scope decision

Included in Agent Pulse with a readable, recently refreshed public-git source.

Jul 29, 04:34 AM GMT+2

Included in Agent Pulse with a readable, recently refreshed public-git source.

Jul 29, 04:34 AM GMT+2
betagouv/api.gouv.fr Excluded, with reason

Watched as the first API Discovery provider catalog, but excluded from Agent Pulse because it belongs to the daily source-brief lane rather than the AI-agent project shelf.

Editorial scope decision
cbouvat/saracroche-ios Excluded, with reason

Watched for an editor-requested public mobile source profile, but excluded from Agent Pulse because Saracroche is a call-filtering application, not an AI-agent or agent-adjacent project.

Editorial scope decision

Included in Agent Pulse with a readable, recently refreshed public-git source.

Jul 29, 04:34 AM GMT+2

Included in Agent Pulse with a readable, recently refreshed public-git source.

Jul 29, 04:34 AM GMT+2

Included in Agent Pulse with a readable, recently refreshed public-git source.

Jul 29, 04:34 AM GMT+2

Included in Agent Pulse with a readable, recently refreshed public-git source.

Jul 29, 04:34 AM GMT+2

Included in Agent Pulse with a readable, recently refreshed public-git source.

Jul 29, 04:34 AM GMT+2
cline/cline Tracked

Included in Agent Pulse with a readable, recently refreshed public-git source.

Jul 29, 04:35 AM GMT+2

Included in Agent Pulse with a readable, recently refreshed public-git source.

Jul 29, 04:35 AM GMT+2

Included in Agent Pulse with a readable, recently refreshed public-git source.

Jul 29, 04:35 AM GMT+2

Included in Agent Pulse with a readable, recently refreshed public-git source.

Jul 29, 04:35 AM GMT+2

Included in Agent Pulse with a readable, recently refreshed public-git source.

Jul 29, 04:35 AM GMT+2

Included in Agent Pulse with a readable, recently refreshed public-git source.

Jul 29, 04:35 AM GMT+2

Included in Agent Pulse with a readable, recently refreshed public-git source.

Jul 29, 04:35 AM GMT+2

Project Ledger

Source map ›
Open the full measured ledger24 included projects
openai/codex Coding agent / Tracked
172 7d commits

openai/codex shows 172 public commits this week, down 101 from the prior week and 6.5 below its current 30-day pace, with tools as the strongest subject signal.

30d 765 vs prior 7d -101 contributors 104 coverage 76 confidence High

Watch whether the cooldown is a pause after heavy work or the start of a quieter cycle.

QwenLM/qwen-code Coding agent CLI / Tracked
284 7d commits

QwenLM/qwen-code shows 284 public commits this week, down 32 from the prior week and 5.4 above its current 30-day pace, with tools as the strongest subject signal.

30d 1,194 vs prior 7d -32 contributors 92 coverage 5 confidence High

Watch whether the cooldown is a pause after heavy work or the start of a quieter cycle.

mastra-ai/mastra Agent framework / Tracked
203 7d commits

mastra-ai/mastra shows 203 public commits this week, down 75 from the prior week and 37.6 below its current 30-day pace, with tools as the strongest subject signal.

30d 1,031 vs prior 7d -75 contributors 104 coverage 1 confidence High

Watch whether the cooldown is a pause after heavy work or the start of a quieter cycle.

google-gemini/gemini-cli Coding agent CLI / Tracked
16 7d commits

google-gemini/gemini-cli shows 16 public commits this week, up 10 from the prior week and 5.5 above its current 30-day pace, with tools as the strongest subject signal.

30d 45 vs prior 7d +10 contributors 11 coverage 54 confidence High

Watch the next tools commits for evidence that the theme is becoming product behavior.

browser-use/browser-use Browser agent framework / Tracked
49 7d commits

browser-use/browser-use shows 49 public commits this week, up 40 from the prior week and 13.3 above its current 30-day pace, with tools as the strongest subject signal.

30d 153 vs prior 7d +40 contributors 10 coverage 0 confidence Medium

Watch whether the subject-proxy work turns into visible release or user-facing control changes.

charmbracelet/crush Coding agent TUI / Tracked
41 7d commits

charmbracelet/crush shows 41 public commits this week, down 47 from the prior week and 11 below its current 30-day pace, with tools as the strongest subject signal.

30d 223 vs prior 7d -47 contributors 23 coverage 14 confidence High

Watch whether the cooldown is a pause after heavy work or the start of a quieter cycle.

OpenHands/OpenHands Coding agent platform / Tracked
49 7d commits

OpenHands/OpenHands shows 49 public commits this week, down 3 from the prior week and 7.2 below its current 30-day pace, with tools as the strongest subject signal.

30d 241 vs prior 7d -3 contributors 34 coverage 1 confidence High

Watch whether the subject-proxy work turns into visible release or user-facing control changes.

langchain-ai/langchain Agent framework / Tracked
31 7d commits

langchain-ai/langchain shows 31 public commits this week, even with the prior week and 1.2 below its current 30-day pace, with ux as the strongest subject signal.

30d 138 vs prior 7d 0 contributors 23 coverage 11 confidence High

Watch whether the subject-proxy work turns into visible release or user-facing control changes.

microsoft/agent-framework Agent framework / Tracked
37 7d commits

microsoft/agent-framework shows 37 public commits this week, down 8 from the prior week and 18.5 below its current 30-day pace, with tools as the strongest subject signal.

30d 238 vs prior 7d -8 contributors 66 coverage 2 confidence High

Watch whether the subject-proxy work turns into visible release or user-facing control changes.

openinterpreter/openinterpreter Coding agent / Tracked
26 7d commits

openinterpreter/openinterpreter shows 26 public commits this week, down 171 from the prior week and 123.1 below its current 30-day pace, with tools as the strongest subject signal.

30d 639 vs prior 7d -171 contributors 88 coverage 2 confidence High

Watch whether the cooldown is a pause after heavy work or the start of a quieter cycle.

4 7d commits

google-antigravity/antigravity-cli shows 4 public commits this week, down 3 from the prior week and 0.2 below its current 30-day pace, with observability as the strongest subject signal.

30d 18 vs prior 7d -3 contributors 6 coverage 3 confidence Medium

Watch the next observability commits for evidence that the theme is becoming product behavior.

shepherd-agents/shepherd Agent runtime / Tracked
0 7d commits

shepherd-agents/shepherd is quiet this week, but still shows 117 commits across the last 30 days.

30d 117 vs prior 7d -5 contributors 6 coverage 2 confidence Medium

Watch whether the subject-proxy work turns into visible release or user-facing control changes.

Coverage Boundary

Read methodology ›

V0.2 reports one dated decision brief from direct public source evidence and keeps git motion as a secondary measurement layer. Community feedback metrics are absent until a reproducible feedback collector ships.

local git commit historycommit subjectsrecent touched file pathspublished article coveragereported direct-source daily insight
not collected: GitHub issue flownot collected: GitHub PR review timingnot collected: Discord feedbacknot collected: X/Twitter feedbacknot collected: Reddit or Hacker News feedbacknot collected: private maintainer backlog