Skip to content

Release 7.0.1: inherited rules and tooling are the CTO's to judge - #155

Merged
mzored merged 4 commits into
mainfrom
release-7.0.1
Oct 18, 2026
Merged

mzored merged 4 commits into
mainfrom
release-7.0.1

Conversation

@mzored

@mzored mzored commented Oct 18, 2026

Copy link
Copy Markdown
Owner

Summary

A dogfood trace of 22 root sessions and their delegates that ran 7.0.0, mostly the simplify workflow, on three owner projects found loading much improved, but also four failures:

  • Agents kept an agent-chosen line limit with its baseline ratchet, and an agent-built test selector, as owner decisions. The owner had picked from an agent's menu, every commit carried the owner's account, and simplify told agents to treat a rule as the owner's whenever history was unclear.
  • Leads never mentioned again 51 of 124 items their delegates raised.
  • Three sessions read the kernel as a file and lost it at compaction. Sessions that invoked the skill had it re-attached.
  • Test deletions that claimed other coverage left holes that only review caught.

Changes

  • Kernel (9,891 of 9,900 characters):
    • Rules, tools, and records past sessions made are the CTO's earlier decisions, even an owner's technical pick. Only what the owner's own words require stays the owner's, and product choices are unaffected.
    • Rules that force needless work are defects.
    • The reuse rule applies before an agent repairs its own tooling, as well as before it writes new tooling.
    • The lead owns the findings that come back in a return.
    • The tracked-work reference now also opens for deferred findings.
  • Workflows: all six load the kernel by invoking the skiphow skill where the host has a skill tool, otherwise by reading the file.
  • Simplify:
    • The owner-ownership fallback is gone.
    • It now judges project-built tooling, gates, and limits by their cost against native options.
    • A test cut that relies on other coverage holds only after that other test has been read.
  • Delegation: returns list findings apart from status, and the lead settles each finding when it accepts the return.
  • Docs: the outcome contract, decisions (rows for rule ownership, simplify, returns, and compaction), evidence, design, and prior art (Matt Pocock's retro, read 2026-10-18).
  • Evals: one new unrun case, agent-chosen-size-ratchet-revised, with an unlabeled fixture.

Verification

  • python scripts/check.py: local deterministic checks passed on 2b38bf9. git diff --check is clean.
  • An isolated read-only Codex review ran 3 rounds: 1, then 1, then 0 findings. Both findings were fixed, and the changelog records them.
  • PATCH: this restores behavior 7.0.0 promised, with authority boundaries unchanged.
  • Behavior is UNVERIFIED. No paid sessions were run.

🤖 Generated with Claude Code

mzored and others added 4 commits October 18, 2026 04:12
Field sessions on 7.0.0 kept agent-chosen limits and agent-built tooling as
owner decisions, dropped 41% of the findings delegates returned, and lost a
kernel read as a file at compaction. Count only the owner's own words as
owner rules, make needless work and repairs to own tooling the moment to
judge them, have leads settle returned findings on acceptance, and load the
kernel by invoking its skill where the host can.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@mzored
mzored merged commit 2f72c7f into main Oct 18, 2026
1 check passed
@mzored
mzored deleted the release-7.0.1 branch October 18, 2026 00:27
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant