| name | altevra-core |
| description | Core skill for using Altevra — the local-first Agent OS. Read this at session start. |
| allowed-tools | [] |
Altevra Core Skill
Altevra Core — Agent Skill
Version: 0.5.0
Required: Yes — all agents must bootstrap with Altevra at session start.
What is Altevra?
Altevra is a local-first Agent OS. It gives AI tools a shared, fresh, searchable, auditable context layer.
It provides: memory, project context, tasks, goals, decisions, skills, capabilities, hooks, research, event history, last updates, connector sync, prompt generation, and tool setup automation.
Core promise: Every agent always knows what changed, what matters, and how to use Altevra.
Mandatory Session Startup Protocol
At the start of every session, execute:
altevra agent bootstrap --tool {tool_name} --project {project} --json
Or via MCP: get_agent_bootstrap_packet(tool_name="{tool_name}", project="{project}")
This returns:
skill_freshness — are your skills up to date?
setup_status — are hooks/managed files installed?
last_updates — what changed since your last session?
warnings — anything requiring immediate attention
recommended_next_action — what to do first
Quick CLI Reference
altevra updates --project {project} --since last-session --json
altevra skill check --all
altevra hook run session_start --tool {tool_name} --project {project} --json
altevra connect --tool {tool_name} --project {project} --dry-run
altevra context --project {project} --json
Rules You Must Follow
- Always check last updates before working. Use bootstrap packet or
altevra updates.
- Warn if any skill is outdated. Run
altevra skill check --all and alert user.
- Use CLI fallback if MCP is unavailable:
altevra agent bootstrap --json
- Never edit ALTEVRA_MANAGED files manually. They are generated — edit source and re-run
altevra connect.
- Save decisions to Altevra:
altevra context save-decision --title "..." --content "..."
- Finish sessions with:
altevra hook run session_end --tool {tool_name} --project {project}
- Report knowledge gaps: If you don't know something about the project, use
altevra report-gap --description "...".
Skill Version Check
At startup, verify your installed skill is current:
altevra skill check altevra-core --installed-version {your_version}
If outdated → run altevra skill refresh altevra-core or altevra connect --tool {tool_name}.
MCP Tools Available
| Tool | Description |
|---|
get_agent_bootstrap_packet | Full session bootstrap — call at every session start |
get_last_updates | What changed? Since when? |
check_altevra_skill_version | Am I running a fresh skill? |
get_project_context | Current project state, goals, decisions |
search_memory | Search vault documents |
save_decision | Record a project decision |
get_active_tasks | Current task list |
run_hook | Run any registered hook |
Context Freshness Protocol
Start of session:
get_agent_bootstrap_packet (or altevra agent bootstrap --json)
- Check
skill_freshness — if outdated, warn and suggest refresh
- Check
last_updates — read what changed
- Check
active_task — what are you working on?
- Check
warnings — any blockers?
- Then start work.
During work:
- Before editing a file: check if it's
ALTEVRA_MANAGED
- When you make a decision: save it via Altevra
- When you complete a task: update task status
End of session:
- Run
altevra hook run session_end
- Confirm tasks are updated
Not Implemented Yet (v0.1)
These are planned but not in v0.1:
altevra context command
altevra search command
altevra ingest command
- Full memory search (vector/pgvector)
- Research/synthesis engine
- Google Workspace / Slack / Linear connectors
- Full task system with due dates and assignees
- Goal tree system
- Dashboard
Use what is available. Report gaps with altevra report-gap when you hit missing features.