v1.2.4

Fix two live routing collisions: plain run-tests verification now stays zero-skill, and coding-agent-team intents prefer controller/worker skills over Microsoft Teams meeting integrations.

v1.2.3

Separate optional router skill/CLI with bounded agent-team guidance. Adds coding-agent-team vs Microsoft Teams routing regression coverage and keeps --max 10 explicit-only.

v1.2.2

Prevents the software-development test boost from loading a debugger for plain verification. Debug and failing-test routes remain eligible.

v1.2.1

Keeps context-mode explicit-only for automatic skill routing, preventing ordinary test and verification tasks from loading its heavy handbook. Also extends workflow verb coverage for compression and context-reduction intents.

v1.1.0 — strict zero-skill routing

What changed

• Returns zero skills for trivial and unrelated prompts.
• Strict automatic routing selects at most one high-confidence skill.
• Explicit $skill-name requests still win.
• Portable agent-skill-route CLI for Codex, Claude Code, Hermes, and Giga/GG Coder.
• Neutral-install and routing tests in GitHub Actions.

Why it matters

The router no longer turns every prompt into hidden context. It adds context only when a relevant workflow skill is worth its token cost.

Verification

20 unit tests pass. Clean GG companion installation passes. Trivial prompts produce an empty router block.

agent-token-saver-skill-router v1.0.7

Smarter security-review routing, tested 10-skill staged controller manifests, and a direct bridge to the full agent-token-saver stack.\n\n- Security + review skills beat generic web/API matches for security audits.\n- Default remains 1-3 active skills; 10 is an explicit phased-controller ceiling.\n- Python stdlib only.\n- 15/15 tests pass.\n\nFull stack: https://github.com/Supersynergy/agent-token-saver

agent-token-saver-skill-router v1.0.4

1.0.4

• Routing fix: a skill matching several intent tokens now outranks a skill that hit one lucky rare name token (coverage multiplier in score, locked by a red/green regression test).

• Instant install: install.sh is now curl-pipeable — no checkout needed:

  curl -fsSL https://raw.githubusercontent.com/Supersynergy/agent-token-saver-skill-router/main/install.sh | bash -s -- claude   # or codex / hermes / ggcoder / opencode / repo / all

• Fresh benchmark (2026-07-12, 458 installed skills): 37,080 → 227 estimated catalog tokens, 99.39% reduction (chars/4 estimate). README now invites third-party benchmark issues — run bench on your library and post your numbers.

agent-token-saver-skill-router v1.0.2

README now leads with Save Your Skill Tokens, Humanlove product framing, proof tables, install paths, Hermes setup, FAQ, and verification. Router now scans GG Coder flat ~/.gg/skills/*.md skills and avoids stopword/substring false positives (e.g. si inside using). Verified with Python 3.14.4: py_compile, unittest (4 tests), bench on 422 skills with 99.33% estimated token reduction, and 5 real route/load checks.

agent-token-saver-skill-router v1.0.1

Adds native GG Coder install target at ~/.gg/skills/agent-token-saver-skill-router.md and removes the local Python 3.11 mise pin. Tested with Python 3.14.4, Codex in tmux, and GG Coder CLI/tmux shell checks. Verification: python3 -m py_compile scripts/agent_token_saver.py; python3 -m unittest discover -s tests -v; install --target ggcoder/codex --dry-run; benchmark reduction ~99.52%.

agent-token-saver-skill-router v1.0.0

Universal adaptive skill router for Hermes, Claude Code, Codex CLI, OpenCode, Cursor/Windsurf, and repo-local agents.\n\nMeasured on this machine: 416 skills scanned; full catalog ~33,454 estimated tokens; router block ~172 estimated tokens; ~99.49% reduction.\n\nVerification: python3 -m py_compile scripts/agent_token_saver.py && python3 -m unittest discover -s tests -v.