| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
The review already ran before every commit. It ran in the context that wrote the diff, which is the part that didn't work. A self-review checks the change against the author's own model of it and can't check the model, so the errors that survive are the ones that were never in the diff: a scope inherited from whoever reported the problem, a blast radius estimated instead of measured.
So the reviewer is now a subagent, on every commit including a one-line one. It gets the diff, a one-line claim of what the change does, and the ticket where one exists. It doesn't get the conversation or my reasoning, because those are the model it exists not to hold. The ticket stays in for the opposite reason: it was written before the work and is the only thing that can contradict my claim about my own diff.
Findings loop back to the same reviewer until it approves, rather than to a fresh one that can't tell an addressed finding from one that never existed. Three rounds, or a finding that recurs after being reported fixed, and it stops for a human instead. The unattended callers park the task rather than wait for an answer nobody is awake to give.
I gave the stance a substantiation floor. An agent told to attack will manufacture findings to satisfy the instruction, and a manufactured finding costs a round and teaches the author to discount the next review.
subagents.md said three separate times not to dispatch work this small, so it now carries an Isolation Override: the size heuristics assume the main thread could do the job equally well, and they lapse when its own context is what makes its answer untrustworthy. Field 2 of the Prompt Contract inverts there. Paste your context in and you hand over the very model you spawned the agent to escape.
I wrote it by running it on itself: three rounds and thirteen findings. Two were things I'd have shipped. Withholding the ticket made my own claim self-certifying, and the override reaffirmed the contract field that would have destroyed the isolation.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
commits.md was 12,800 tokens, larger than the next four rules combined, and it loaded in every session in every project whether or not anything was being published. Most of it is procedure that only applies when a commit or PR is in play.
I split it by blast radius rather than by size. What stays always-loaded is what must hold whether or not I'm publishing, and where a violation is permanent and reaches other people: author identity, the no-AI-attribution ban, the generated-document byline rule, and the public-artifact content-scope rules. That core is 1,027 words, about 2,342 tokens.
What moved into the publish skill is everything about how a message gets written, reviewed, approved, and published. The pre-flight reconcile, the code-review gate, the draft and voice and approval gate, conventional-commit format, Voice and Focus, PR description structure, the three review shapes, merge strategy, and the pre-commit checklist.
The failure modes are why the line falls there. If the skill doesn't load I don't know the publish flow and have to be told, which is visible and recoverable. I don't silently commit with AI attribution, because that guard never moved. Only the recoverable half is on the skill-triggering bet.
I verified the split by using it. The skill registered mid-session and I invoked it to publish this commit, and it loaded with the flow intact.
Content is conserved: 5,561 words in, 5,898 across both files, the difference being the frontmatter and the pointer I added to the core. I repointed five cross-references in voice, review-code, inbox.org, and no-approvals.org that named sections which moved.
One thing I left alone. The skill is a single 4,871-word file, and both posts argue a long skill should split internally rather than sit as one blob. It loads on demand now, which is the win worth taking today. Splitting it further is its own change.
|
| |
|
|
|
|
|
|
|
|
| |
Every org document an agent writes carried `#+AUTHOR: Craig Jennings & Claude`. No template stamps that line. Agents copy it from a neighboring file, so one stray header propagates through everything generated afterward.
My own repos tolerate the co-author line. Employers whose policy is that work product carries employee names alone do not. An `#+AUTHOR:` line survives conversion into docx, a wiki page, or a PDF that reaches a customer.
I rewrote the header to `Craig Jennings` across the workflows, templates, specs, and design docs. The rule now lives in commits.md, so the next generated document starts correct rather than inheriting the mistake.
Archived session logs keep their original headers as a record of what happened. The two Codex-authored design docs keep their byline, because Codex wrote them and relabeling would be a false attribution rather than the removal of one.
|
| |
|
|
|
|
|
|
| |
inbox.org's auto mode regains its "run this batch next?" ask, now chaining into work-the-backlog as an explicit second step after routing: the eligibility query over the queued batch, file-only, paging off, cap 1. Startup and wrap-up still never execute.
The no-approvals speedrun lands as the named preset: an explicit ordered list run under autonomous-commit + always-push + paging-on, every approval front-loaded into the seven-step pre-flight. Any phrase containing "speedrun" routes to the preset, with disambiguation notes in no-approvals.org and the index. The finer Q&A mechanics land with Phase 4.
I scoped the chain's task set to the queued batch rather than all of todo.org. The ask is "run this batch next?", and a batch-yes running an unrelated higher-priority task would be surprising.
|
| |
|
|
|
|
|
|
|
|
|
|
| |
artifact budgets
Two patterns kept failing in practice despite being documented (#40 praise asymmetry and the #38 terse cut), so I made the walk verifiable and closed the content gap behind tangled review text.
The high-recurrence set (#13, #37, #38, #40, #42) now gets per-pattern attestation receipts. The anti-AI audit runs after the terse pass so the audited text is the text that ships. Short personal-mode artifacts get a compact output format, and a write-back step puts the voiced text in the file the publish flow posts from.
Four patterns are new: #42 finding stems (one claim per sentence in review findings), #43 single-sentence paragraph cadence, #44 parenthetical asides, #45 declarative register marker. #37 exempts verdict formulas. #40 covers verification narration. #13 and #33 carry the self-discipline framing. A per-artifact budgets table makes terse a checkable budget instead of an adjective.
The profile gains paired entries with the approved worked examples, and commits.md plus no-approvals.org drop hardcoded pattern counts so the next addition doesn't re-drift them.
|
|
|
Pearl independently built its own no-approvals workflow (handoff in this
morning's inbox) and asked rulesets to take the best from both. That's
cross-project signal that the workflow earns a place at the template tier.
Promoted from =.ai/project-workflows/= to =claude-templates/.ai/workflows/=
with a mirror under =.ai/workflows/=. INDEX entry added under "Tools and
meta" listing the full set of trigger phrases.
Merged from pearl's draft: the prominent "What's Suspended" / "What Stays
On" split (same content as the prior "Contract" section, easier to scan),
wider trigger phrases (the project-only version had fewer), the
mode-resets-when-Craig-switches-topics guard, an explicit destructive-action
carve-out as its own bullet rather than buried in the "real question"
section, and a subagent-review-gate callout.
Kept from the rulesets version: the Session Log update emphasis between
items, the real-question include/exclude lists, and the don't-auto-wrap-up
guard.
A History section at the bottom records the merge for future archaeology.
|
