Skip to content

Releases: Ancienttwo/Fable-agents

Release list

v1.1.0

v1.1.0 Latest
Latest

Choose a tag to compare

View all tags
@Ancienttwo Ancienttwo released this 05 Jul 12:32
v1.1.0
d9a627b
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

[1.1.0] - 2026-07-05

Changed

  • fast-worker — now opens with a machine-readable RESULT: DONE/PARTIAL/BLOCKED line, pastes real verification output, labels unverified claims [inferred]/[unverified], and states explicit default boundaries (no commit/push/PR unless ordered).
  • deep-reasoner — now opens with RECOMMENDATION: <one sentence> — confidence: HIGH/MEDIUM/LOW, labels evidence tri-state, and makes its recommend-only / no-side-effects role explicit. Kept on all tools; a local tools: allowlist (e.g. ["Read","Grep","Glob","Bash","WebSearch","WebFetch"]) is a documented opt-in, not the default, because static allowlists block per-machine MCP tools.
  • Re-running the installer on a machine with v1.0.0 copies reports [conflict] for these two agents — diff and take the bundled version.

Fixed

  • install.sh — agent definitions now install globally into $CLAUDE_CONFIG_DIR/agents (default ~/.claude/agents) instead of per-project (fix landed on main after v1.0.0; first released here).
Assets 2
Loading

v1.0.0

v1.0.0

Choose a tag to compare

View all tags
@Ancienttwo Ancienttwo released this 05 Jul 10:44
v1.0.0
28af2ea
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Portable installer for the Fable 5 model-routing setup: fast-worker (Sonnet 5) for execution, deep-reasoner (Opus 4.8) for architecture research, gatekeeper (Opus 4.8) as the acceptance and ship gate, and Codex as an independent peer engineer.

Install

git clone https://github.com/Ancienttwo/Fable-agents.git
cd Fable-agents
bash scripts/install.sh --project /path/to/your/project

Or paste to any Claude that can run commands on your machine:

Install the model-routing setup from https://github.com/Ancienttwo/Fable-agents — clone the repo, run bash scripts/install.sh --project <path-to-my-project>, then report the per-layer results. Never overwrite anything on [conflict].

Why

One Max (20x) subscription, one day: 902M tokens across 4,911 requests at a 96.1% cache-hit rate — $731 in API-equivalent usage. Fable answered only 492 requests (10%, pure orchestration) while Sonnet carried 2,924 execution requests and Opus 705 judgment/gate runs. The two weekly pools drain in step — Fable 32% vs all-models 26% — instead of the Fable pool burning out while the shared pool idles.

Highlights

  • Acceptance and ship gate: gatekeeper reviews the diff against the goal, runs the project's real verification, and returns a PASS/FAIL/BLOCKED ship recommendation — ship actions only on an explicit orchestrator execution order (#1)
  • Idempotent installer: project agents + global CLAUDE.md routing section + codex plugin check + headless smoke tests; never overwrites on [conflict]
  • Bilingual README (English / 简体中文)
  • Usable as a Claude Code skill under ~/.claude/skills/

Full changelog: CHANGELOG.md

Loading