Repository navigation
Fable Mode 2.0.0 — config-driven spend, Claude 5 era overhaul - #8
Merged
Merged
Conversation
Add .claude/fable/CONFIG.md (subagent model/effort, fleet, votes) read by skills at invocation and honored by all 7 workflows via a drift-checked shared config block; new /fable-config skill with value/light/max presets. Installers preserve user CONFIG.md on --update (CI-tested both platforms). Rewrite FABLE.md, GUIDE.md, README, and agent reporting style for the leaner Claude 5 generation: plain language, lead with the outcome, value over token burn. New SVG banner; drop stale PNGs; bump to 2.0.0. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XWSx33Mi2bXpR6w75r3SY2
The dogfood fable-review run (4 finders, 3 skeptics per finding, 19 agents) confirmed 5 findings and refuted none: - Shipped CONFIG.md had votes: 3 uncommented, short-circuiting the fleet vote presets (light=1/standard=3/max=5) — votes now ships commented, and the /fable-config presets re-comment it instead of forcing a count. - fable-review never scaled its finder pool by fleet — light now runs 2 of 4 dimensions, announced. - /ultra hardcoded votes: 3 into review args, overriding user config — it now defers to the config's fleet. - The checker's includes() presence test let a config block with an edited anchor line drop out of the drift comparison while still passing — the required-block check now demands a full twin-regex match (negative-tested). - README's light-fleet agent math corrected to what the code does. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XWSx33Mi2bXpR6w75r3SY2
settings.json holds repo-local permission grants (currently just the checker command), so both installers now skip it like settings.local.json and it is gitignored — .claude/** otherwise ships verbatim to targets. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XWSx33Mi2bXpR6w75r3SY2
…template New pack file .claude/fable/settings.template.json pre-approves the Workflow and Agent tools — the harness-level half of the doctrine's standing authorization. Installers copy it to the target's .claude/settings.json only when none exists; an existing file is never modified (a hint prints the two entries instead), and --update never touches it. Disclosed in SECURITY.md, covered in README install and uninstall, and CI now asserts materialization, the tracked+1 file count, and that updates preserve a user-edited settings.json on both platforms. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XWSx33Mi2bXpR6w75r3SY2
Subagent thinking has no separate switch — subagent_effort is the honest control, now stated in CONFIG.md and the /fable-config skill. The fleet note also gives exact per-preset agent counts instead of adjectives. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XWSx33Mi2bXpR6w75r3SY2
creativeskyai
marked this pull request as ready for review
August 8, 2026 02:25
|
Bugbot is not enabled for your account, so this pull request was not reviewed. Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs. |
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.
What changed
A full 2.0 pass over every layer, adding one new layer: Config (
.claude/fable/CONFIG.md). The file sets the defaults every fable command runs with —subagent_model(ships asopus),subagent_effort(ships asmedium),fleet(light/standard/max), and optionalvotes/ round / lead-cap overrides. Skills read it at invocation and pass it asconfigin workflow args; all 7 workflows apply it via a new drift-checked shared config block (model/effort on every agent call, fleet-scaled pools, votes, and caps, each announced withlog()). A new skill,/fable-config(11th), shows settings, sets keys, or appliesvalue/light/maxpresets, and prints the harness commands for main-agent model/effort/thinking that a pack cannot set itself.Install-time behavior:
fable/CONFIG.mdso--updatenever overwrites user tuning..claude/fable/settings.template.json) is materialized as the target's.claude/settings.jsonon first install only, pre-approving the Workflow and Agent tools so fleet launches don't prompt. An existing settings file is never modified (the installer prints the two entries to add instead),--updatenever touches it, and deleting it opts out. Disclosed in SECURITY.md and the README install/uninstall sections..claude/settings.jsonis never shipped (basename-skipped in the copy loop, gitignored in this repo).Doctrine (
FABLE.md),GUIDE.md, README, and agent reporting were rewritten leaner and plainer for the Claude 5 generation, with a value-over-token-burn framing (recommended driver: Fable 5 at low effort, or Opus 5; subagents on Opus at medium). New SVG banner; stalebanner.png/review-pipeline.pngremoved (listed in the changelog for--updateusers); version bumped to 2.0.0.Why
The pack predates the Claude 5 generation's leaner Claude Code system prompts and Anthropic's updated steering/context-engineering guidance. Fleet size and spend were only steerable by prompt wording, which the model had to infer; users wanted an explicit dial for how many subagents run, on what model, at what effort — plus documented low-effort/value presets, since lower effort with structural verification keeps benchmarking faster and cheaper than higher effort without it.
How it was verified
node tools/check-workflows.cjspasses (7 workflows, 7 agents, 11 skills; new config-block twin + required-block checks included)settings.json, one import line), idempotent second run (0 copied),--updaterestores tamperedFABLE.mdbyte-identical while preserving a user-editedCONFIG.mdandsettings.jsonCHANGELOG.mdupdated and.claude/fable/VERSIONbumped to 2.0.0 (checker enforces the match)fable-reviewworkflow ran over this branch through the new config path; its 5 confirmed findings were fixed in a follow-up commit on this PR🤖 Generated with Claude Code
https://claude.ai/code/session_01XWSx33Mi2bXpR6w75r3SY2