Repository navigation
Trim the supervisor's playbook; RED's compile block names the missing symbol (run 26) - #43
Merged
Merged
Conversation
… symbol (run 26) SKILL.md is re-read on every supervisor turn. The sections a normal run does not need (Codex/ OpenCode notes, the budget re-plan route and breach classes, review blind spots, the script list, transcripts, worktree hygiene) move to skills/aloop/reference.md behind one-line pointers; OpenCode installs copy it beside the skill (aladdin-cli test). Run 26 (wordfreq): 13/13, $0.95; the playbook load fell from 25.0k to 18.8k tokens. Run 26's cost came from a RED loop: the block said only "add mod-common shims" after RED had to revert a src/ skeleton. gate-red-verify now names each symbol the tests cannot find and the shim to add in tests/common/mod.rs (conformance case, red before the change). Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014j4wLin3oJTtnvjE9Lq2VV
adeelahmad
pushed a commit
that referenced
this pull request
Oct 18, 2026
…not .aladdin/ SKILL.md's layout tree listed `sprintN/` below `<project>/.aladdin/` (run data), and live run 24's haiku supervisor put the whole sprint in `.aladdin/sprint1/` (#41 made the gates refuse it). The tree now nests sprintN/ under docs/agents/ and draws .aladdin/ as run data only. scripts/test/skill-layout.test.py checks the tree and holds SKILL.md to a 50 KB budget so #43's trim does not grow back. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_015QxyVf3NgwsgRyYpBgDDTC
This was referenced Oct 18, 2026
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.
Requested by Adeel · project thread
Before: the supervisor re-read the whole 58k-char playbook on every turn, including sections a normal Claude Code run never uses. And when RED's tests did not compile, the block said only "add mod-common shims", without the symbol or the shim's shape.
After: the playbook is 16% shorter. The Codex/OpenCode notes, the budget re-plan route and breach classes, review blind spots, the script list, transcripts and worktree hygiene live in
skills/aloop/reference.md, read on demand from one-line pointers; OpenCode installs copy it beside the skill. The RED compile block now names each symbol the tests cannot find and the shim to add, for examplepub fn count_words(…) -> … { todo!() }intests/common/mod.rs, marked// agentic:shimand called throughmod common;.How: a section move with pointers in SKILL.md, the
budgetmessages and the planner persona pointing at the new file, an installer copy for OpenCode, and a rustc error parse (E0425/E0412/E0422/E0423, E0432, E0599) ingate-red-verify.Evidence: live run 26 (wordfreq): 13/13, 16m01s, $0.95. The playbook load fell from 25.0k to 18.8k tokens, so about 6k fewer tokens are re-read on every supervisor turn, and the supervisor never needed the reference file. The run's cost was set by a RED loop instead (5 of 7 blocks): its mandate allowed a
src/skeleton, the gate made it revert, and the tests then called a symbol that no longer existed. That loop is what the new block message fixes. Tests: conformance 63/63 (the new §6.3 case was red before the change),aladdin-cli.test.py5/5 (reference.md installed, red before),make test-hosts pycheck json version-check shellcheck eval-validateOK,check-tracking47 requirements,claude plugin validate --strictpasses.Generated by Claude Code