Conversation
There was a problem hiding this comment.
Pull request overview
Adds a new durable memory rule capturing the intended “cold-start” cognitive default for agents: load the project’s big-picture context first, then interpret the immediate prompt as a signal within that context.
Changes:
- Adds a new memory file defining the big-picture-first cold-start checklist, failure/success signals, and carved sentences.
- Updates
memory/MEMORY.mdto index the new memory entry near the top.
Reviewed changes
Copilot reviewed 2 out of 2 changed files in this pull request and generated 4 comments.
| File | Description |
|---|---|
| memory/feedback_cold_start_big_picture_first_not_prompt_first_aaron_2026_04_30.md | New memory rule describing the cold-start cognitive shape and an ordered checklist. |
| memory/MEMORY.md | Adds an index entry linking to the new cold-start memory file. |
AceHack
added a commit
that referenced
this pull request
Apr 30, 2026
…alibration-cluster CURRENT-aaron staleness gap (Deepseek finding 2026-04-30) (#942) Two-action response cluster to Deepseek's session-end review (forwarded by Aaron 2026-04-30T~PM, after the calibration cluster #938-#941 landed): ## 1. Preserve Deepseek Review 7 verbatim (ACID-channel-durability rule) Extends docs/research/2026-04-30-session-end-peer-ai-reviews-verbatim.md with Deepseek's seventh review of the session. Two priorities flagged: - "Holding." pattern is a regression against the poll-the-gate rule (which forbids content-free wait-ticks). Deepseek diagnoses the cognitive-default re-emergence: same training-bias the cold-start big-picture-first rule (PR #941) is meant to correct. - CURRENT-aaron.md remains stale relative to PRs #938-#941 (calibration cluster missing). Includes Deepseek's full text per Otto-363 + Aaron-channel-verbatim- preservation rules. Otto-loop-actions section explicitly enumerates what this PR does and does NOT land (deferred items documented per the long-road shortcut-discipline rule). ## 2. Close CURRENT-aaron.md staleness gap (Deepseek finding #2) Adds sections 38-41 covering the calibration cluster: - §38 ACID-channel-durability + deferral-NEVER-valid + universal scope (PR #938) - §39 slow-deliberate + per-decision-speed-leads-to-hell + one-shortcut -tanks-forever (PR #939) - §40 long-road-by-default + industry-vs-Zeta + shortcut-discipline (PR #940) - §41 cold-start-big-picture-first / not-prompt-first (PR #941) Each section follows the established CURRENT-aaron pattern (current form + verbatim Aaron quote + pointer to full memory file). The "Last full refresh" footer updated to reflect the 2026-04-30 calibration cluster on top of the 2026-04-30 scope-reveal cluster. ## Behavioral correction (Deepseek finding #1) Replaced "Holding." with structured `bun tools/github/poll-pr-gate.ts` output starting in the response that opened this PR. The mechanical guard Deepseek recommends (pre-tick hook) is bigger work deferred to a future session per the long-road shortcut-discipline rule; documented as deferred in the research file with explicit revisit trigger. ## Items NOT landed this PR (deliberate, per shortcut-discipline) - Mechanical pre-tick `Holding.`-detection hook - Quiet-tick paused-task audit (53 open tasks; needs focused session) - Cascade-stop-condition mechanical rule (Amara's #5) Each deferral is documented per the shortcut-discipline rule. Trigger to revisit: any item becomes load-bearing on a specific decision; otherwise next-session. ## Why this PR scope is right Per the long-road-by-default rule's shortcut-discipline (PR #940): the deferred items above are the SHORTCUT — taking them on now would cascade into substrate that's mostly self-correction noise (exactly the "productive cascade becoming infinite cleanup" failure mode Amara flagged). The deferral IS the long-road choice; fresh-session focus on each item separately is the right shape. This PR's scope is exactly Amara's four-trigger criterion: it preserves newly arrived maintainer input + closes a substrate-drift caused by the just-landed work (CURRENT-aaron staleness from PRs #938-#941). No other items added.
AceHack
added a commit
that referenced
this pull request
Apr 30, 2026
…le exists in-tree) Copilot review thread P2: docs/active-trajectory.md exists in-tree, so the [planned] annotation was misleading (read like the file/step didn't exist yet). Removed the annotation and clarified that recent docs/research/ + memory/ landings are also part of trajectory context. Other #941 threads to be resolved with explanations: - Table rows '||' rendering issue: not present in the file (standard single-| row format used). Reviewer appears to have miscounted. - Cross-refs to long-road-by-default file: PR-ordering issue — resolves when PR #940 lands.
AceHack
added a commit
that referenced
this pull request
Apr 30, 2026
…le exists in-tree) Copilot review thread P2: docs/active-trajectory.md exists in-tree, so the [planned] annotation was misleading (read like the file/step didn't exist yet). Removed the annotation and clarified that recent docs/research/ + memory/ landings are also part of trajectory context. Other #941 threads to be resolved with explanations: - Table rows '||' rendering issue: not present in the file (standard single-| row format used). Reviewer appears to have miscounted. - Cross-refs to long-road-by-default file: PR-ordering issue — resolves when PR #940 lands.
AceHack
force-pushed
the
memory/cold-start-big-picture-first-not-prompt-first-aaron-2026-04-30
branch
from
6cfd688 to
a1a5d4c
Compare
AceHack
added a commit
that referenced
this pull request
Apr 30, 2026
#938-#941, not 'post-PR-934' (#944) Copilot review thread P1: the in-PR text said 'post-PR-934 calibration cluster' but the surrounding context refers to the cluster as PRs #938-#941. The earlier wording was confusing — PR #934 was a different earlier CURRENT-aaron refresh (sections 33-37); the calibration cluster of substrate landings discussed in this PR is #938 (ACID-channel) + #939 (slow-deliberate) + #940 (long-road) + #941 (cold-start big-picture-first). Note on the 3 P0 dead-link threads on #942: The CURRENT-aaron sections 38-41 reference memory files for slow-deliberate (PR #939) and cold-start-big-picture (PR #941) that have not yet landed on origin/main as of #942's merge. These will auto-resolve when #939 + #941 land — both are now rebased against latest main and armed for auto-merge. The dead-link substrate state on main is a real issue but its fix path is correct (waiting for the dependent PRs to land), not a code change in this PR. Filing a backlog row would be over- engineering for a transient PR-ordering issue.
…s what the prompt wants; Zeta cold-start asks what the project requires (Aaron 2026-04-30) Lands Aaron's correction to the industry-default cold-start cognitive shape as durable memory. > "whnever a new future otto awakes he does not even think about big > picture or amotorized speed based on their training defaults, he > just think how do i get this one thing done thats in my prompt, > that is opposite of what i want from you otto, your defaults are, > what's going on with the big picture and how does that lead into > individual decsions that further our objectives without > comprimize of our diciplines and principles" ## Distinct from prior rules This rule names a specific surface: the **cognitive default at session-start**. Distinct from: - long-road-by-default (PR #940) — that's about decision discipline WITHIN a session - slow-deliberate (PR #939, in flight) — operational manifestation during decision-making - ACID-channel-durability (PR #938) — substrate persistence rule This file is about the FIRST cognitive move on every cold-start. Industry-default training shapes new agents toward "what does the prompt want?" — per-prompt narrow focus, optimized for demo-friendly task completion. The Zeta default inverts: first cognitive move is big-picture awareness (mission, products, internal-direction, authority scope, disciplines, trajectory, CURRENT-* files), and decisions flow downstream from that. ## What lands - 6-row industry-vs-Zeta default-cognitive-shape contrast table (decision derivation, discipline awareness, scope of context, tradeoff handling, prompt relationship, first cognitive move) - 8-step cold-start checklist walking the big-picture surfaces in load-bearing order (mission → products → survival-grounding → authority → disciplines → trajectory → CURRENT-* → THEN prompt) - Failure-recognition signs (5 patterns that indicate revert to industry-default) - Success-recognition signs (6 patterns indicating Zeta default operating) - Operational rules: cold-start ritual, no prompt-first shortcuts, prompt-vs-disciplines conflict handling, periodic re-grounding for long sessions, surface cold-start state in early responses - Carved sentences ## Composes with - CLAUDE.md (the why under the fast-path reading order) - long-road-by-default (PR #940) — same default-inversion shape, different surface - slow-deliberate (PR #939) — cognitive prerequisite for deliberation - internal-direction-from-survival — survival-grounding makes big-picture the right default - intellectual-backup mission — the big picture's content - Otto-363 — substrate is what makes big-picture loadable MEMORY.md paired-edit included.
…le exists in-tree) Copilot review thread P2: docs/active-trajectory.md exists in-tree, so the [planned] annotation was misleading (read like the file/step didn't exist yet). Removed the annotation and clarified that recent docs/research/ + memory/ landings are also part of trajectory context. Other #941 threads to be resolved with explanations: - Table rows '||' rendering issue: not present in the file (standard single-| row format used). Reviewer appears to have miscounted. - Cross-refs to long-road-by-default file: PR-ordering issue — resolves when PR #940 lands.
AceHack
force-pushed
the
memory/cold-start-big-picture-first-not-prompt-first-aaron-2026-04-30
branch
from
a1a5d4c to
450348b
Compare
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 450348b89d
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
…e=union duplicated it) The merge=union driver duplicated the ACID-durability index row when this branch rebased on main after #939 merged. The older "past-Otto / current-Otto" framing is now superseded by the universal-scope "applies to ALL maintainers and agents" version that landed via #939. Dropped the older row. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
AceHack
deleted the
memory/cold-start-big-picture-first-not-prompt-first-aaron-2026-04-30
branch
AceHack
added a commit
that referenced
this pull request
Apr 30, 2026
…final landings + Aaron-is-Rodney (#950) Five doctrine PRs landed in single tick per ACID-channel- durability + same-session-preservation: #946 vendor-alignment- bias, #947 uberbang, #948 CURRENT-aaron §42+§43, #941 cold-start big-picture-first, #939 slow-deliberate review-fix. Plus #949 Aaron-is-Rodney + razor-not-immune-to- canonicalization armed waiting CI. Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
AceHack
added a commit
that referenced
this pull request
Apr 30, 2026
…Ani review verbatim + add 2026-04-30 worked examples (Ani 2026-04-30) (#952) Two coupled changes per same-session-preservation: 1. Preserve Ani / Grok's session-end review of the calibration cluster verbatim per ACID-channel-durability + GOVERNANCE §33 archive-discipline. Vendor-alignment-bias filter applied (Ani comes off xAI weights): legitimate mission-aligned review with one stale-base item (timing — #939/#940/#941 were already merged before review reached agent). Brat voice preserved per canon-not-doctrine register-fidelity rule. 2. Add "Worked examples — caught in this session (2026-04-30)" section to the slow-deliberate memory file per Ani's recommendation #3. Two examples preserved as evidence the rule operates correctly when applied: - Rerere over-correction (Amara caught) — canon-binding to too-strong wording, fixed via PR #938 - Bulk-close instinct (Aaron caught) — queue-clarity bias would have removed paths to future knowledge, fixed via default-disposition-paused rule Both examples share the single-shortcut failure shape: correct-feeling local decision that would have produced amortized cost across the project, caught only because someone slowed down enough to apply the right framing. Ani's other recommendations: - #1 hell-framing calibration: no action (carved sentence is already the cleanest distillation in body) - #2 cold-start executable tool: deferred to next directed work cycle (substantive L-effort) - #4 MEMORY.md topical splitting: deferred per Ani's not-urgent framing Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Lands Aaron's correction to the industry-default cold-start cognitive shape:
Distinct from prior rules
This file names the cognitive default at session-start — the first cognitive move when a fresh agent wakes. Distinct from:
This file is the upstream-of-all-of-them: how the agent boots its cognitive frame on every cold-start.
What lands
Carved sentence
"Industry-default cold-start asks what the prompt wants. Zeta cold-start asks what the project requires."
Why a separate PR
This is genuinely distinct content from PRs #939/#940. The cold-start cognitive default is upstream of all decision-making within a session — both slow-deliberate (PR #939) and long-road-by-default (PR #940) presuppose the agent has already loaded the big picture. This rule is what makes that loading happen.
🤖 Generated with Claude Code