Trailmark

[English](./README.md) · [Русский](./README.ru.md)

[](https://opensource.org/licenses/MIT) [](./manifest.yml) [](./docs/SKILLS_MAP.md) [](https://github.com/bearded-illirian/trailmark/stargazers) [](https://github.com/bearded-illirian/trailmark/issues)

Trailmark

**An artifact-first agent framework for driving multi-step engineering tasks through a disciplined skill chain — one block at a time, one commit at a time, every decision written down.**

Every step your AI agent takes writes a file. Every file registers in a local database. Every **block** *(one unit of work inside a task)* closes with a **report** *(the markdown summary written after execution)*. Nothing lives only in chat.


**Find any artifact — decision, plan, report, note — in 3-4 clicks.**


No more "hey Claude, remind me what we did in that task last week." Open Flow UI, click the task, see every file the block produced. Reviewable, searchable, permanent.

---

What Trailmark solves

Three pains every Claude Code developer hits — and how Trailmark structurally prevents each.

1. Session amnesia — "what did we decide yesterday?"

Chat sessions are ephemeral. Every new window = context reset. Two weeks later you can't recall why you chose approach X over Y, or what alternatives you considered.

**Trailmark's answer:** every decision writes a file — `flow-first-N.md` (landscape), `library-fi