Optimize setup context budget: tiered skills, deferred reads, pruned …#13
Open
Optimize setup context budget: tiered skills, deferred reads, pruned …#13
Conversation
…re-reads
Three changes to reduce setup's context consumption by ~2,400 lines:
1. Tiered skill generation (Step 9): Classify 16 skills into three
tiers — A (copy: learn, tasks), B (frontmatter-only: reflect,
reweave, stats, graph, refactor), C (DOMAIN substitution: seed,
pipeline, ralph, rethink, next, validate, remember, reduce, verify).
Tiers A+B skip body reasoning entirely; Tier C does mechanical
string replacement only. Corrects plan misclassification of reduce
and verify (both have {DOMAIN:xxx} patterns, moved from B to C).
2. Defer upfront reference reads: Trim mandatory reads from 8 files
(~2,268 lines) to 3 (kernel.yaml, interaction-constraints.md,
vocabulary-transforms.md). Defer use-case-presets to Step 3c
(matched preset only), personality-layer to Step 3d (only if
signals detected). Inline failure-modes vulnerability matrix into
Step 3f. Drop three-spaces.md (covered by Step 2) and
conversation-patterns.md (consult on ambiguity only).
3. Eliminate redundant re-reads: Remove 6 of 14 ops/derivation.md
re-reads (Steps 4, 5, 5f, 5g, 6, 11 — small files from info
already in context) and 2 interaction-constraints.md re-reads
(Steps 3b, 3e — already loaded upfront).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…4 and 5g - failure-modes.md: "not needed" annotation contradicted line 446 which still said "consult." Changed to scoped deferred read (HIGH-risk sections only) with explicit instruction to use prevention patterns for Common Pitfalls. - Step 4 (identity.md): restored re-read of derivation.md + conditional personality-layer.md re-read. Step 3 is the largest generation task; without the re-read anchor, personality dimensions drift out of attention. - Step 5g (manual/): restored re-read of derivation.md before 7 manual pages. Prevents vocabulary drift across a long multi-file generation. Co-Authored-By: Claude Opus 4.6 <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
1 participant
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.
…re-reads
Three changes to reduce setup's context consumption by ~2,400 lines:
Tiered skill generation (Step 9): Classify 16 skills into three tiers — A (copy: learn, tasks), B (frontmatter-only: reflect, reweave, stats, graph, refactor), C (DOMAIN substitution: seed, pipeline, ralph, rethink, next, validate, remember, reduce, verify). Tiers A+B skip body reasoning entirely; Tier C does mechanical string replacement only. Corrects plan misclassification of reduce and verify (both have {DOMAIN:xxx} patterns, moved from B to C).
Defer upfront reference reads: Trim mandatory reads from 8 files (~2,268 lines) to 3 (kernel.yaml, interaction-constraints.md, vocabulary-transforms.md). Defer use-case-presets to Step 3c (matched preset only), personality-layer to Step 3d (only if signals detected). Inline failure-modes vulnerability matrix into Step 3f. Drop three-spaces.md (covered by Step 2) and conversation-patterns.md (consult on ambiguity only).
Eliminate redundant re-reads: Remove 6 of 14 ops/derivation.md re-reads (Steps 4, 5, 5f, 5g, 6, 11 — small files from info already in context) and 2 interaction-constraints.md re-reads (Steps 3b, 3e — already loaded upfront).