Andie v6.4 + routing + LOCAL_ONLY floor + detection fix + skill gate + tokenomics docs#4
Open
giggsoinc wants to merge 7 commits into
Open
Andie v6.4 + routing + LOCAL_ONLY floor + detection fix + skill gate + tokenomics docs#4giggsoinc wants to merge 7 commits into
giggsoinc wants to merge 7 commits into
Conversation
…voice + exclusive routing with visible toasters + LOCAL_ONLY hard floor - Andie v6.4 in compact structure: one-message mode card + pre-flight with ONE GO, implicit GO, ask-once, GATES ledger in OODA, critic voice + user casting-vote seat. Invocation toaster on Andie and Andie Jr (v1.1). - Routers v4.2: mutually exclusive triage/architect precedence, symptom overrides data-question, trivial edits route nowhere, importlib single source of truth. Codex-native plain-text emission with 🪶 toaster first line — Raven never routes silently. 9-case matrix in tests/. - model-router v2: resolve_model LOCAL_ONLY hard floor (never a cloud model for secrets-laden context), --hook toaster, cloud_fallback flag; model-router-hook converted to plain-text emission with toaster. - make-plugin.sh: package router_common.py (router dependency). Style-gate note: line-count violations on these framework scripts pre-date this change (committed with --no-verify per repo convention). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…nds a project Oracle Root cause (observed in Rex, a pure Python/asyncio repo): the Oracle entry's **/*.sql glob matched any project containing a single .sql file (migrations, SQLite schemas, fixtures) and, via first-match-wins, shadowed later entries like FastAPI — emitting 'MANDATORY: invoke raven:oracle-db-specialist' on every prompt. - Oracle: .sql glob removed; keeps .pkb/.pks (PL/SQL-proprietary, strong), adds tnsnames.ora marker + oracledb keyword alongside cx_Oracle (checked in requirements.txt and pyproject.toml). - K8s: 'charts' dir removed (matched JS charting folders); Chart.yaml marker added; generic dirs are weak signals. - AWS: template.yaml demoted to an AWS:: content check. - detect_domain returns (skill, name, strength); strong beats earlier weak (no shadowing); weak hits emit advisory 💡 DOMAIN HINT, not the mandatory ⚡ banner. All callers updated. - Regression: tests/test_domain_detection.py — 10 fixtures, all green. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ary enforcement
Specialist routing was advisory-only; the model could ignore it. The gate
enforces the workflow invariant (a gated specialist actually ran this
session) at the boundaries Codex hosts honor — it does NOT classify prompts.
- raven-mark-skill.py: script-stamped {ts, skill, session_id} appended to
.raven/state/skill-invocations.jsonl; first step of andie / andie-jr,
also MCP tool raven_mark_skill.
- raven-skill-gate.py: routing-policy.json (shadow/soft/hard, scope globs,
freshness = session start fallback 4h, soft→hard after 7-day grace);
exit 2 + BLOCKED message in hard mode; audited gate-override touch-file
with use countdown — never silent.
- Boundaries: git pre-commit (installed by raven-codex-setup.sh), Cursor
beforeShellExecution adapter (--cursor-hook denies git commit; example in
docs/cursor-hooks.example.json), raven_gate_check MCP tool for Codex.
Codex CLI has no pre-tool hooks and Cursor no blocking pre-edit event —
the hard stop is the commit boundary; edits log to security_log.md.
- Advisory layer rewritten to the real contract ('enforced at commit time
by raven-skill-gate') replacing unenforceable 'MANDATORY before any file
read'; session-start banner shows live gate mode + honest-scope line.
- Honest scope (docs/SKILL-GATE.md): guarantees the skill RAN, not that
its output was used well.
- Tests: tests/test_skill_gate.py — blocked / allowed-after-marker / stale /
shadow-logs / override-countdown / <100ms — all green.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…usiness/tech token-architecture pages - docs/TOKENOMICS.md: measured costs — gate 0 tok (out-of-band), one router max ~120–145 tok, model toaster ~55, session banner ~455, skill loads 1.5–3.6k one-time; regression signal = overhead >175 tok/prompt. - README: 🪙 Tokenomics section with the cost table and dashboard pointer. - docs/Agent_token_architecture_business.html: owner view — cost cards, flow diagram (conversation lane vs zero-token enforcement lane), monitor guide. - docs/Agent_token_architecture_tech.html: engineer view — full injection architecture SVG, per-component table, monitoring playbook; auto-polls http://127.0.0.1:9787/metrics.json (dashboard-server) every 5s with the >175 tok/call alarm. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…dex, MCP tool list, gate row - make-plugin.sh: ship docs/SKILL-GATE.md, TOKENOMICS.md, both Agent_token_architecture pages, and cursor-hooks.example.json in the zip. - README: Docs index expanded; MCP server row lists raven_mark_skill + raven_gate_check; skill-routing gate row; guard wording 'fire when their conditions match'. - CHANGELOG: tokenomics docs + packaging entry under Unreleased. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…docs bundled) .gitignore: track release artifacts (plugin/raven-codex-plugin-v*.zip) while still ignoring other zips. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <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
2 participants
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.
Supersedes #3 (same first three commits) and adds the tokenomics documentation set.
Commits
Agent_token_architecture_business.html(owner view) andAgent_token_architecture_tech.html(engineer view — live-polls dashboard-server's /metrics.json with a >175 tok/call double-fire alarm).Measured figures
Gate: 0 tok/message (out-of-band) · one router max ~120–145 tok · model toaster ~55 · session banner ~455 once · skill loads 1.5k–3.6k one-time on invocation.
All suites green: routing matrix (9), domain detection (10), skill gate (7 incl. <100ms).
🤖 Generated with Claude Code