[lydakis]

Requested by George · project thread

One sentence group in lead.md's principle paragraph, added after #394's "weigh it against the rest of the person's work", plus a README line. No code.

What changes

• Plans become work. What the person said they will do or want done (a release, a deadline, a follow-up) is work the lead holds. When nothing owns it, the lead starts it, or sets it up with a trigger for when it is due, and says so in a line.
• Undoable steps only. Nothing goes out in the person's name unasked; that rule is already in lead.md.
• Memory sets how far. How far to go unasked is the person's call: the lead follows what their memory says. When they welcome or push back on something done unasked, the lead saves that. This is the same feedback loop Sonny: leads and tasks give straight feedback, coach repeated habits and weigh what sources report #394 added for nudges.

Grounding

Behavior	Existing pattern
Surfacing commitments that fell through the cracks	Microsoft's Briefing email listed "outstanding commitments, requests, and follow-ups that might have fallen through the cracks", found in Outlook (support page; paused January 2023). It surfaced commitments but never acted on them.
Acting only on stated goals	OpenAI Model Spec: "never overstep and attempt to autonomously pursue goals in ways that aren't directly stated" (spec). A plan the person stated is directly stated.
Feedback that sets how proactive to be	ChatGPT Pulse "remembers your feedback for next time" (OpenAI)

Starting the work, not only surfacing it, goes past what these products ship. That is why it is limited to undoable steps and set by memory.

Not evaluated yet

A commitment in the scripted eval would lead to real triggers, and on macOS those land in the real launchd domain even with a fake HOME. This will join the planned two-day eval once that cleans up its triggers.

---

Generated by Claude Code

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-30T01:38:05.939419Z	c5496a5	PR opened
🔒 Security Review	✅ Completed 2026-10-30T01:39:54.603646Z	c5496a5	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[lydakis]

The runtime job failed on tests.test_served test_a_turn_interrupted_mid_setup_stops_it_and_starts_no_task. The cause is a race in the test: it reads /proc/PID/stat after checking that the file exists, and the process can exit between those two steps (ProcessLookupError). This PR changes only lead.md and the README, so it can't cause that failure. rust, macos and GitGuardian all passed. This session can't re-run CI (403), so I merged on the flake rule. No fix for the race exists yet.

---

Generated by Claude Code