Command reference¶

The governance core (init, doctor, close, log, evidence, handoff) is own logic and works standalone, with Python only. The rest does a transparent shell-out to external tools (optional interop) and shows their output without hiding errors.

Command	What it does	Type
`tramalia menu`	interactive menu	core
`tramalia init`	generate the convention (idempotent)	core
`tramalia doctor [--fix]`	diagnose tools and how to install them	core
`tramalia detect`	detect the stack and applicable gates	core
`tramalia close [--task --agent --reviewer --allow-fail --engram]`	closing ritual: gates → evidence → handoff (enforcement)	core ★
`tramalia log`	audit trail of closes	core ★
`tramalia evidence [--task --engram]`	create the closing evidence pack	core
`tramalia handoff [--task --agent --reviewer --engram]`	multi-agent handoff	core
`tramalia gates`	run the quality gates	interop (mise)
`tramalia context`	generate derived memory (token-saver)	interop (repomix + stdlib)
`tramalia sync [--to ...]`	propagate AGENTS.md to other agents	interop (rulesync)
`tramalia skills [sync\\|list]`	clone/update skills from their repos	interop (git)
`tramalia update`	update everything	interop (mise + copier + skills)
`tramalia mcp`	start the MCP façade	core (+ mcp SDK)

close — the governance ritual¶

The flagship command. In one step: it runs each gate (mise run <gate>), writes their output into the evidence pack, generates the handoff, and blocks the close if a gate fails (unless you pass --allow-fail with the exception noted in risks.md).

tramalia close --task TASK-001 --agent codex --reviewer claude

It works standalone: if mise is missing, it does not invent a result — it records in the pack that the gates did not run as a documented exception, and still leaves evidence + handoff.

Each close writes metadata.json (task, agent, reviewer, timestamps, exit codes and an honest status: passed / blocked / passed_with_exceptions / no_gates). The raw *-output.txt files are the official evidence; no derived artifact (e.g. Headroom compression) may replace them.

log — the audit trail¶

Reads each close's metadata.json and lists the closes (newest first) with their status and agent. It's the verifiable history of agentic work on the repo.

doctor¶

Classifies requirements into bootstrap (mise/git/uv), stack (node/dotnet…) and feature/gate (semgrep/sqlfluff/lighthouse…), bothering you only with what applies to your project. --fix delegates to mise install when mise is present.

init¶

Generates idempotently (never overwrites existing files): a single AGENTS.md, CLAUDE.md (@AGENTS.md), docs/ai/ with the curated rules (DB, security and UX/UI gates), a stack-tailored mise.toml, .mcp.json with Serena, and .tramalia/ (config + skills).

evidence and handoff¶

Tramalia's two own pieces for traceability:

evidence creates .tramalia/evidence/<date>-<task>/ with summary, files-changed (reads git diff), commands, each gate's output, risks, rollback and next-steps.
handoff appends a structured entry to docs/ai/07-handoff-agentes.md.

sync¶

rulesync convert --from agentsmd --to copilot,cursor,cline --features rules. It excludes Claude/Codex (they already read AGENTS.md natively). Configurable with --to.

mcp — the façade (level 1)¶

Exposes the convention as native MCP tools: project_status, get_agent_rules, get_failed_attempts, get_current_task, doctor, record_handoff, build_evidence, build_context. Wire it in .mcp.json:

{
  "mcpServers": {
    "tramalia": { "command": "tramalia", "args": ["mcp"] }
  }
}