From 9a55fd069bba5e35990ad1aa197d4a0453b31f5b Mon Sep 17 00:00:00 2001 From: Affaan Mustafa Date: Sun, 29 Mar 2026 00:14:17 -0400 Subject: [PATCH] fix(skills): harden token budget advisor skill --- AGENTS.md | 4 +-- README.md | 6 ++-- manifests/install-modules.json | 1 + skills/token-budget-advisor/SKILL.md | 52 +++++++++++++++++----------- 4 files changed, 37 insertions(+), 26 deletions(-) diff --git a/AGENTS.md b/AGENTS.md index 750414de..e195b696 100644 --- a/AGENTS.md +++ b/AGENTS.md @@ -1,6 +1,6 @@ # Everything Claude Code (ECC) — Agent Instructions -This is a **production-ready AI coding plugin** providing 28 specialized agents, 125 skills, 60 commands, and automated hook workflows for software development. +This is a **production-ready AI coding plugin** providing 29 specialized agents, 133 skills, 60 commands, and automated hook workflows for software development. **Version:** 1.9.0 @@ -141,7 +141,7 @@ Troubleshoot failures: check test isolation → verify mocks → fix implementat ## Project Structure ``` -agents/ — 28 specialized subagents +agents/ — 29 specialized subagents skills/ — 125 workflow skills and domain knowledge commands/ — 60 slash commands hooks/ — Trigger-based automations diff --git a/README.md b/README.md index 8d3b5bd3..c28048f1 100644 --- a/README.md +++ b/README.md @@ -212,7 +212,7 @@ For manual install instructions see the README in the `rules/` folder. /plugin list everything-claude-code@everything-claude-code ``` -✨ **That's it!** You now have access to 28 agents, 125 skills, and 60 commands. +✨ **That's it!** You now have access to 29 agents, 133 skills, and 60 commands. --- @@ -273,7 +273,7 @@ everything-claude-code/ | |-- plugin.json # Plugin metadata and component paths | |-- marketplace.json # Marketplace catalog for /plugin marketplace add | -|-- agents/ # 28 specialized subagents for delegation +|-- agents/ # 29 specialized subagents for delegation | |-- planner.md # Feature implementation planning | |-- architect.md # System design decisions | |-- tdd-guide.md # Test-driven development @@ -1085,7 +1085,7 @@ The configuration is automatically detected from `.opencode/opencode.json`. |---------|-------------|----------|--------| | Agents | ✅ 28 agents | ✅ 12 agents | **Claude Code leads** | | Commands | ✅ 60 commands | ✅ 31 commands | **Claude Code leads** | -| Skills | ✅ 125 skills | ✅ 37 skills | **Claude Code leads** | +| Skills | ✅ 133 skills | ✅ 37 skills | **Claude Code leads** | | Hooks | ✅ 8 event types | ✅ 11 events | **OpenCode has more!** | | Rules | ✅ 29 rules | ✅ 13 instructions | **Claude Code leads** | | MCP Servers | ✅ 14 servers | ✅ Full | **Full parity** | diff --git a/manifests/install-modules.json b/manifests/install-modules.json index 8b6e175a..c35578ff 100644 --- a/manifests/install-modules.json +++ b/manifests/install-modules.json @@ -407,6 +407,7 @@ "skills/ralphinho-rfc-pipeline", "skills/regex-vs-llm-structured-text", "skills/search-first", + "skills/token-budget-advisor", "skills/team-builder" ], "targets": [ diff --git a/skills/token-budget-advisor/SKILL.md b/skills/token-budget-advisor/SKILL.md index 4f863e6f..019c4366 100644 --- a/skills/token-budget-advisor/SKILL.md +++ b/skills/token-budget-advisor/SKILL.md @@ -5,10 +5,13 @@ description: >- how much depth/tokens to consume — BEFORE responding. Use this skill when the user wants to control token consumption, adjust response depth, choose between short/long answers, or optimize their prompt. - TRIGGER when: "token budget", "token count", "token usage", "token limit", + TRIGGER when: "token budget", "response token budget", "token count", + "token usage", "response length", "response depth", "brief answer", + "short answer", "detailed answer", "full answer", "respuesta corta vs larga", "cuántos tokens", "ahorrar tokens", "responde al 50%", "dame la versión corta", "quiero controlar cuánto usas", - "75%", "100%", "at 25%", "at 50%", "at 75%", "at 100%", "exhaustive", or any variant where the user wants + "75%", "100%", "at 25%", "at 50%", "at 75%", "at 100%", + "give me the full answer", or any variant where the user wants to control length, depth, or token usage — even without mentioning tokens. DO NOT TRIGGER when: user has already specified a level in the current session (maintain it) or the request is clearly a one-word answer. @@ -28,25 +31,16 @@ Intercept the response flow to offer the user a choice about response depth **be **Do not trigger** when: user already set a level this session (maintain it silently), or the answer is trivially one line. -## Workflow +## How It Works ### Step 1 — Estimate input tokens -Use the calibration tables below to estimate the prompt's token count mentally. +Use the repository's canonical estimation guidance from `skills/context-budget`. -**Chars-per-token by content type:** +- Prose-first prompts: `input_tokens ≈ word_count × 1.3` +- Code-heavy or mixed prompts: `input_tokens ≈ char_count / 4` -| Content type | Chars / Token | -|-------------------|---------------| -| English natural | ~4.0 | -| Spanish natural | ~3.5 | -| Code | ~3.0 | -| JSON | ~2.8 | -| Markdown | ~3.3 | - -Formula: `input_tokens ≈ char_count / chars_per_token` - -For mixed content, use the dominant type's ratio. +For mixed content, prefer the code-heavy estimate as the conservative default. ### Step 2 — Estimate response size by complexity @@ -104,10 +98,10 @@ If the user already signals a level, respond at that level immediately without a | What they say | Level | |----------------------------------------------------|-------| -| "25%" / "short" / "brief" / "tldr" / "one-liner" | 25% | -| "50%" / "moderate" / "normal" | 50% | -| "75%" / "detailed" / "thorough" / "complete" | 75% | -| "100%" / "exhaustive" / "everything" / "full" | 100% | +| "1" / "25%" / "short answer" / "brief" / "tldr" / "one-liner" | 25% | +| "2" / "50%" / "moderate detail" / "balanced answer" | 50% | +| "3" / "75%" / "detailed answer" / "thorough explanation" | 75% | +| "4" / "100%" / "exhaustive" / "everything" / "full answer" | 100% | If the user set a level earlier in the session, **maintain it silently** for subsequent responses unless they change it. @@ -115,7 +109,23 @@ If the user set a level earlier in the session, **maintain it silently** for sub This skill uses heuristic estimation — no real tokenizer. Accuracy ~85-90%, variance ±15%. Always show the disclaimer. +## Examples + +### Triggers + +- "Give me the brief answer first." +- "How many tokens will your response use?" +- "Respond at 50% depth." +- "I want the full answer." +- "Dame la version corta." + +### Does Not Trigger + +- "Explain OAuth token refresh flow." (`token` here is domain language, not a budget request) +- "Why is this JWT token invalid?" (security/domain usage, not response sizing) +- "What is 2 + 2?" (trivially short answer) + ## Source Standalone skill from [TBA — Token Budget Advisor for Claude Code](https://github.com/Xabilimon1/Token-Budget-Advisor-Claude-Code-). -Full version includes a Python estimator script for exact counts: `npx token-budget-advisor`. +The upstream project includes an optional estimator script, but this ECC skill intentionally stays zero-dependency and heuristic-only.