Preview. These maps are our reading of public sources and haven't been checked with the projects' maintainers yet.
Trail-code (formerly Graft)
A codebase map plus team 'learnings' fed to Claude Code, Cursor, Codex and other agents through hooks, to cut search tokens and tool calls.
github.com/trailhq/Trail-CodeChanged direction
Show HN 2026-08-14 as NanoNets/Graft; the repo now redirects to trailhq/Trail-Code, which says 'Graft is now Trail-code' and adds 'learnings'. Last push 2026-10-26; ~10.8k stars. (checked 2026-10-25)
In Skills and plugin packs for coding agents
Where it stands apart
- Is it actively maintained? active, but changed direction (pivot, fork, or rename) (0% of 19 others)
- How is it mainly sustained? company with a paid product, tier, or ads (22% of 18 others)
- Which shared standard or format does it mainly build on? an agent's plugin or hooks format (Claude Code plugins and hooks, OpenCode plugins) (11% of 19 others)
- What is most of it, whatever it's packaged as? code that runs alongside the agent (a plugin, hooks, or an MCP server) (11% of 19 others)
- How big is it? one skill or tool with one job (21% of 19 others)
- What work is it for? one technical area (frontend, security, codebase search, vision) (16% of 19 others)
- Where does its know-how mainly come from? data mined at scale (posts, designs, benchmarks) (11% of 19 others)
- Does it publish evidence that it makes agents better? yes, evals or benchmarks with numbers (11% of 19 others)
- Does it send usage data to its maker? on by default, with an opt-out (11% of 19 others)
What it shares with most of the community
- How is the main thing licensed? permissive open source (MIT/Apache/BSD) (84% of 19 others)
- Can it run entirely on your own device or infrastructure, without the maker's servers? yes (95% of 19 others)
- Do you need an account to use it? no account (95% of 19 others)
- Who is it for? developers (79% of 19 others)
- Does it need an outside service beyond the agent and its model? no (74% of 19 others)
Nearest neighbors
- frontend-design-pro 59% of 17 shared answers match
- HumanLayer show-me 53% of 17 shared answers match
- OpenCode Senses 53% of 17 shared answers match
Every answer
| Question | Answer | Source |
|---|---|---|
| Is it actively maintained? | stands apart active, but changed direction (pivot, fork, or rename) | source 2026-10-26 |
| How is the main thing licensed? | permissive open source (MIT/Apache/BSD) | source 2026-10-26 |
| How is it mainly sustained? | stands apart company with a paid product, tier, or ads Trail (trailhq.com) sells a 'company brain' platform for agents; Trail-code is its open-source codebase layer. | source 2026-10-26 |
| Can it run entirely on your own device or infrastructure, without the maker's servers? | yes The graft map is built locally with tree-sitter, 'No model, no API key'. | source 2026-10-26 |
| Do you need an account to use it? | no account | source 2026-10-26 |
| Who is it for? | developers | source 2026-10-26 |
| Which shared standard or format does it mainly build on? | stands apart an agent's plugin or hooks format (Claude Code plugins and hooks, OpenCode plugins) Feeds its codebase map to agents through each agent's hooks. | source 2026-10-26 |
| Which coding agents does it give install steps for? | 6 or more named agents, or any agent via a cross-agent installer (npx skills) | source 2026-10-26 |
| What is most of it, whatever it's packaged as? | stands apart code that runs alongside the agent (a plugin, hooks, or an MCP server) | source 2026-10-26 |
| How is it mainly installed? | a package manager or installer CLI (npm, npx skills, pip) | source 2026-10-26 |
| How big is it? | stands apart one skill or tool with one job | source 2026-10-26 |
| What work is it for? | stands apart one technical area (frontend, security, codebase search, vision) Codebase search and context. | source 2026-10-26 |
| Does it need an outside service beyond the agent and its model? | no | source 2026-10-26 |
| How does it get the agent to follow it? | instructions plus scripts or checks the agent is told to run Hooks inject map/learnings; agent is told to run trail check (exit 1 on conflict) before commit or PR. | source 2026-10-26 |
| Where does its know-how mainly come from? | stands apart data mined at scale (posts, designs, benchmarks) Indexes your own code; closest option is data at scale. | source 2026-10-26 |
| Does it publish evidence that it makes agents better? | stands apart yes, evals or benchmarks with numbers 162 runs: 46% fewer tool calls, 42% fewer uncached input tokens; SWE-bench Verified 33/50 vs 27/50 with Claude Sonnet 5. | source 2026-10-26 |
| Does it send usage data to its maker? | stands apart on by default, with an opt-out README badge: 'telemetry anonymous, opt-out'; TELEMETRY.md in repo. | source 2026-10-26 |
Sources
This page was derived from:
- GitHub · github.com/trailhq/Trail-Code · read 2026-10-25
- Web page · trailhq.com/ · collected 2026-10-26
RSS for this project · Its own release feed
Are you a maintainer? Claiming and correcting projects is coming soon.