Skip to content

docs: fix README Deep-lane routing rule to match the implementation - #1

Merged
stid merged 1 commit into
mainfrom
docs/readme-lane-table
Jul 25, 2026
Merged

docs: fix README Deep-lane routing rule to match the implementation#1
stid merged 1 commit into
mainfrom
docs/readme-lane-table

Conversation

@stid

@stid stid commented Jul 25, 2026

Copy link
Copy Markdown
Owner

The contradiction

The README lane table and the actual triage logic disagree on when work routes to Deep.

Source Rule
README.md:38 (before this PR) "4+ signals, or architecture / parallel surfaces / EVAL / irreversibility"
skills/triage/SKILL.md:59 "4+ → Standard, UNLESS a risk signal is among them → Deep. Pure soft-signal accumulation caps at Standard."
rules/lcd.md:34 same risk-gated rule
docs/why-lcd.md:139 "4+ signals route to Deep only if a risk signal is among them"
docs/vs-sdd.md:94 "Hard triggers + risk signals; soft-signal accumulation caps at Standard"

The README is the only file carrying the pre-risk-gate rule — it looks like a leftover from
before routing became risk-gated. Everything else in the tree agrees.

This matters more than a typo: a reader who scores 4 soft signals (files, surfaces, local
architecture, cold-pickup) and follows the README lands in the full spec → plan → tasks → test-gen → red-green → audit pipeline, when the implementation deliberately caps that at
Standard. The front door inverts the methodology's own core claim on exactly the case the
three-lane model exists to keep light.

Changes

  • README.md — lane table now states the risk-gated rule, with one sentence under the signal
    list spelling out the cap (file count alone never buys the heavy pipeline).
  • README.md — "Once published to a marketplace:" → "From the marketplace:" (LCD is published).
  • CHANGELOG.mdUnreleased entry, per CONTRIBUTING.

Testing

  • tests/run.sh — 9/9 pass
  • claude plugin validate . — PASS
  • No eval rows: this touches neither skill, command, nor rule prose (docs only), so the
    benchmark is not implicated.

Also reviewed, no drift found

Skills (6), commands (8), bundled rules (7 + lcd.md), hooks (3 matchers, scripts present),
why-lcd.md's eight pillars, doc/example cross-links, the no-dependencies claim
(lcd-hooklib.sh degrades jq → python3 → node → empty, fail-open), quickstart.md's
onboarding description vs skills/onboard/SKILL.md, monorepos.md keys vs
templates/lcd-conventions.md, and a full sweep for residual pre-rename strings (none).

🤖 Generated with Claude Code

https://claude.ai/code/session_014CfxbekXCXLz6EeqckNq6b

The lane table read "4+ signals, or architecture / parallel surfaces / EVAL
/ irreversibility" — the pre-risk-gate rule. skills/triage/SKILL.md,
rules/lcd.md, docs/why-lcd.md (pillar 5) and docs/vs-sdd.md all specify
risk-gated routing: any hard trigger goes Deep, otherwise 4+ signals go Deep
only when a risk signal (irreversibility or multi-session cold-pickup) is
among them, and pure soft-signal accumulation caps at Standard.

A reader following the README would over-route accumulated-file-count work
into the full pipeline — the case the three-lane model exists to keep light.

Also drops the stale "Once published to a marketplace" conditional.

Docs only; no skill/command/rule prose touched, so no eval rows needed.
Verified: tests/run.sh 9/9, claude plugin validate PASS.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014CfxbekXCXLz6EeqckNq6b
@stid
stid merged commit 00ac449 into main Jul 25, 2026
3 checks passed
@stid
stid deleted the docs/readme-lane-table branch July 25, 2026 06:35
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