tokenbender/agent-guides
tokenbender/agent-guides has 369 GitHub stars and 29 forks. GitHot measured 0 stars in 24 hours. Last checked 2026-11-18 00:42 UTC. Last push 2026-07-23.
Measured star growth
- in 24h
- 0
- in 7 days
- Collecting history
- in 15 days
- Collecting history
- in 1 month
- Collecting history
Net star changes over overlapping periods, not daily totals. Missing measurements are left blank. Stars measure attention, not software quality.
About
one page guides that i let my subscribed/customised agents consume to perform actions
Developer ToolsREADME.md
OriginalAgent Guides
Field guides and a production skill collection for Claude Code, current with the July 2026 ecosystem: commands merged into skills, native subagents with worktree isolation, lifecycle hooks, path-scoped memory, and the Agent Skills open standard.
The repo tracks the ecosystem as it moves. Last full audit: July 2026 against
code.claude.com/docsand METR's published time-horizon data.
📚 Guides
| Guide | What it covers |
|---|---|
| Agent Skills | SKILL.md anatomy, full frontmatter, dynamic context injection, context: fork subagent execution, precedence, distribution |
| Custom Commands | /commands quick start: arguments, frontmatter, built-in command map |
| Subagents | .claude/agents/, background-by-default execution, worktree isolation, persistent memory, orchestration patterns |
| Hooks | Deterministic guardrails: event table, matchers, decision JSON, five production recipes |
| Memory | CLAUDE.md, .claude/rules/ path scoping, auto memory, AGENTS.md interop, what-goes-where table |
| Code Search | Natural-language codebase search, memory files that improve it, custom search skills |
| Conversation Search | Session history: JSONL storage format, /resume picker, jq/rg power recipes |
Which surface when: facts → memory · procedures → skills · isolated or parallel work → subagents · hard enforcement → hooks · external systems → MCP.
🛠️ Skills Collection
claude-skills/ — 13 skills in standard SKILL.md layout, dependencies (references, templates, scripts) included. Install one or all:
cp -r claude-skills/multi-mind ~/.claude/skills/ # one
cp -R claude-skills/*/ ~/.claude/skills/ # all
| Skill | What it does |
|---|---|
| multi-mind | Fans judgment questions out to 4-6 decorrelated specialist subagents with adversarial cross-pollination rounds |
| capability-horizon-estimator | Answers "can the model do this, and how long will it take" with METR time-horizon math: logistic success curve, doubling-rate extrapolation, best-of-N, wall-clock calculator |
| epistemic-libido | Filters and ranks papers, posts, and ideas for high-entropy mechanistic insight and underpriced leverage — separates fertile results from hype |
| audit-sft-data-quality | Audits SFT datasets against the behavior they teach: contracts, contamination, lineage, train/review/reject dispositions |
| tokenbending | Experiment discipline: deterministic contract vs fuzzy choices, preserve-before-delete, stop-on-first-failure |
| worklog | Issue-led atomic work journal: one issue per unit, comment ledger, closeout checklist |
| paper-ocr-notes-pipeline | Paper PDF/URL → grounded OCR + teachable notes.md, with identity gate and revision passes |
| pdf-ocr-feedback | High-accuracy OCR via self-scoring rubric and Maj@K consensus voting |
| technical-writing-workflow | Full technical writing pipeline: ontology tables, outline contracts, rigor and anti-slop passes, audit scripts |
| blade-and-contract | Provocative but defensible rhetoric: lead with the blade, anchor with the contract |
| playwright | Browser automation from the terminal via playwright-cli, wrapper script included |
| manim-math-explainer | Manim Community math scenes: style rules, templates, render loop |
| x-thread-reader | Full X/Twitter threads through twitter-thread.com with fallbacks |
Try the newest one
# Can a 12-hour-horizon model do a 4-hour task?
python3 claude-skills/capability-horizon-estimator/scripts/estimate.py --h50 719 --task-min 240
# SINGLE ATTEMPT: 70.6% VERDICT: LIKELY WALL-CLOCK EST: ~48 min
🔧 Scripts
extract-claude-session.py— exports any session from~/.claude/projects/JSONL transcripts to markdown or cleaned JSONL.
Contributing
Add a skill (format: Agent Skills guide), fix a fact, or open a PR with a workflow worth sharing.
License
See LICENSE.