Skip to content

backlog(B-0158): adopt .claude/rules pattern + carved-sentences-only constraint (Aaron 2026-05-01)#1133

Merged
AceHack merged 3 commits intomainfrom
backlog/b0158-claude-rules-pattern-carved-sentences-only-aaron-2026-05-01
May 1, 2026
Merged

backlog(B-0158): adopt .claude/rules pattern + carved-sentences-only constraint (Aaron 2026-05-01)#1133
AceHack merged 3 commits intomainfrom
backlog/b0158-claude-rules-pattern-carved-sentences-only-aaron-2026-05-01

Conversation

@AceHack
Copy link
Copy Markdown
Member

@AceHack AceHack commented May 1, 2026

Summary

Aaron 2026-05-01 directional input:
"i mean honestly the way ../no-copy-only-learning-agents-insight does .claude/rules seems like the way to go and you are only allowed carved statments in the rules compression is key."

Files B-0158 capturing the design + carved-sentences-only constraint.

Outcome-driven framing

This row is a trajectory (per parallel-trajectories discipline). Outcomes solved:

  1. Reduce always-on context-load — CLAUDE.md 23.7k + GOVERNANCE.md 40.4k (AT proven ceiling) + AGENTS.md 19.2k = 107.5k chars.
  2. Reduce goldfish-ontology recurrence — per-rule canonical homes make pre-author grep cheap.
  3. Carved-sentences-only compression — Aaron's tighter constraint than sibling.
  4. DORA lead-time — every session-wake pays the always-on tax.

External anchor

Sibling repo ../no-copy-only-learning-agents-insight ships this pattern at production scale (proven incident: "Large CLAUDE.md will impact performance 50.1k > 40.0k"). Mechanical anchor, not Aaron-as-anchor.

Three phases

  • Phase 1 (P1) — directory + 3 pilot rules (verify-before-deferring / future-self-not-bound / substrate-or-it-didn't-happen) + carved-sentences linter as B-0153 candidate class.
  • Phase 2 (P2) — bulk migration of standing rules from CLAUDE.md + GOVERNANCE.md.
  • Phase 3 (P3) — parallel-harness consolidation (@<path> references to .cursor/rules/<rule>.mdc).

Dependencies

depends_on: [B-0006, B-0153] per the forward-compat schema landed in the 2026-05-01 backlog-hygiene extension.

Test plan

  • markdownlint clean (npx markdownlint-cli2 EXIT=0)
  • depends_on schema follows the 2026-05-01 extension
  • composes_with linked memos exist on disk
  • outcome-driven framing leads (per Aaron's 2026-05-01 trajectories correction)

🤖 Generated with Claude Code

Copy link
Copy Markdown

Copilot AI left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

Adds a new P1 backlog trajectory row (B-0158) proposing a migration to a .claude/rules/<rule>.md layout and enforcing a carved-sentences-only compression constraint, aimed at reducing always-on context load and improving rule discoverability.

Changes:

  • Introduces backlog row B-0158 with outcomes, external anchor, and phased migration plan.
  • Captures dependencies and composition links to related backlog rows/memory entries.
  • Defines Phase 1 acceptance criteria including a carved-sentences linter concept.

AceHack added 3 commits May 1, 2026 14:45
…nces-only constraint (Aaron 2026-05-01)

Aaron 2026-05-01 directional input:
*"i mean honestly the way ../no-copy-only-learning-agents-insight
does .claude/rules seems like the way to go and you are only
allowed carved statments in the rules compression is key."*

Outcome-driven framing (parallel-trajectories per
docs/active-trajectory.md):

1. Reduce always-on context-load — CLAUDE.md 23.7k +
   GOVERNANCE.md 40.4k (AT proven 40k ceiling) +
   AGENTS.md 19.2k = 107.5k always-on chars across our
   four surfaces. Sibling-repo experience indicates
   measurable performance impact above 40k per-file.
2. Reduce goldfish-ontology recurrence — per-rule
   canonical homes make pre-author grep cheap.
3. Carved-sentences-only compression — Aaron's tighter
   constraint (sibling repo carries prose bodies; we
   carve to blade-shape).
4. DORA lead-time — every session-wake pays the
   always-on tax.

External anchor: sibling repo's claude-md-lean.md
documents the proven incident (Large CLAUDE.md will
impact performance 50.1k > 40.0k); mitigation was the
.claude/rules/<rule>.md split. We are 23.7k on
CLAUDE.md today + 40.4k on GOVERNANCE.md — ceiling on
the horizon, not theoretical.

Three phases:
- Phase 1 (P1) — directory + 3 pilot rules
  (verify-before-deferring, future-self-not-bound,
  substrate-or-it-didn't-happen) + carved-sentences
  linter as B-0153 candidate class.
- Phase 2 (P2) — bulk migration of standing rules
  from CLAUDE.md + GOVERNANCE.md.
- Phase 3 (P3) — parallel-harness consolidation
  (@<path> references to .cursor/rules/<rule>.mdc).

Carries depends_on: [B-0006, B-0153] per the
forward-compat schema landed in the 2026-05-01
backlog-hygiene extension.
…+ replace PR-number with stable framing (Copilot 6 threads)
@AceHack AceHack force-pushed the backlog/b0158-claude-rules-pattern-carved-sentences-only-aaron-2026-05-01 branch from 5d4d398 to 4061340 Compare May 1, 2026 18:46
@AceHack AceHack merged commit a6b6eb8 into main May 1, 2026
21 checks passed
@AceHack AceHack deleted the backlog/b0158-claude-rules-pattern-carved-sentences-only-aaron-2026-05-01 branch May 1, 2026 18:53
AceHack added a commit that referenced this pull request May 1, 2026
…ules + carved-sentences-only canonical) + #1130 final 2 threads drained (#1138)
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.

2 participants