In an interactive terminal, `truthmark init` shows a numbered multi-select for Codex, Claude Code, GitHub Copilot, OpenCode, Antigravity, and Cursor. Select zero or more hosts; choose `none` for a host-neutral, CLI-only repository.
For scripts and CI, repeat `--platform` to replace the selected host set without prompting:
```bash
truthmark init --platform codex --platform cursor
truthmark init --json
```
`--json` never prompts. On a first noninteractive run with no `--platform`, initialization stays host-neutral; later reruns retain the platforms already saved in `.truthmark/config.yml`.
Now try the most common adoption path: document one existing behavior from code and tests. In your AI coding host, ask the installed workflow:
```text
/truthmark-document document the implemented session timeout behavior across src/auth/session.ts and tests/auth/session.test.ts
```
After that, users should not normally invoke Truth Sync directly. Keep coding through your AI host; the installed repository instructions tell the agent to run relevant tests and perform the Truth Sync review before handoff when functional code changes. You review the resulting code diff plus truth-doc diff.
If you only want CLI validation and do not want host-specific AI workflows yet, select `none` and continue with `truthmark check`; you can rerun `truthmark init` later to select platforms.
AI coding agents are incredible at writing code fast. But this speed creates a dangerous new failure mode: **the repository's story drifts from reality.**
**Truthmark** installs a Git-native workflow layer into your repository. It fixes the part of AI development that usually breaks: keeping documentation aligned with code after the first draft.
Truthmark is not a one-shot docs generator. It is an ongoing truth-doc curation loop that keeps human-facing docs small, owned, evidence-backed, and reviewable as agents keep changing code.
* **✅ Trust Through Verification:** AI work becomes easier to trust because behavior-changing work includes a human-reviewable truth-doc decision or diff.
*`truthmark init` - create or refresh configuration, select zero or more platforms, and install routing, truth-doc scaffolds, and selected AI-host instructions.
Truthmark installs repo-local instructions for supported AI coding hosts such as Codex, Claude Code, GitHub Copilot, OpenCode, Antigravity, and Cursor.
The normal loop is simple:
1. Ask your agent for a code change, or ask it to document existing behavior.
2. The installed instructions tell the agent when to test, when to update truth docs, and when to stop for human review.
3. You review ordinary Git diffs: code plus any truth-doc changes.
The user-started agent requests are intentionally few:
*`/truthmark-document` - document existing implemented behavior from code and tests.
*`/truthmark-realize` - implement code from existing truth docs.
*`/truthmark-check` - audit repository truth.
Truth Sync is not the usual way to start work; it is the finish-time review after functional code changes.
Truth Structure is not a day-to-day command; it repairs routing or ownership only when that blocks the work.
Most AI tools can draft documentation. Truthmark keeps repository truth curated after the draft, after the next code change, and after the doc starts getting too broad.
For command-by-command usage, surface comparisons, supported platform details, configuration, routing, Portal, and examples, read the [Truthmark User Guide](docs/user-guide.md).
It is not a hosted service, MCP server, vector database, hidden memory layer, CI enforcement product, or autonomous code rewrite engine. It helps repository truth stay visible; it does not replace tests, code review, or human judgment.
