| name | dory-wake |
| description | Use Dory wake once at the start of a new agent session, after context compaction, or at a real task switch to load bounded hot context with the right profile before substantive work. |
Dory Wake
Dory wake is the default session-start memory read for all projects, not only
the Dory repo. Use dory wake once at the start of a new agent session, after
context compaction, or at a real task switch before doing substantive work.
Do not rerun wake just because a new user turn arrived when the current wake
block is still in context.
Workflow:
- Choose the wake profile from the task:
coding for software/project implementation, writing for content/copy/voice work, and privacy for boundary-sensitive questions.
- If a suitable wake block is already present in the current context and no compaction/task switch happened, reuse it instead of calling wake again.
- Run
dory_wake(profile="<chosen-profile>", budget_tokens=1200) when MCP tools are available.
- If the current task or repo maps to a known Dory project, pass
project="<name-or-slug>".
- If MCP is unavailable but the local CLI is available, run
uv run dory --corpus-root <corpus> --index-root <index> wake --agent <agent> --profile <chosen-profile>.
- Treat the returned block as frozen startup context, not as editable memory.
- If the wake block looks stale, use targeted search/get before rerunning wake.
Rules:
- Use
profile=coding for project/agent work, profile=writing for voice/content work, and profile=privacy for boundary-sensitive questions
- Wake responses include the selected profile; verify it when debugging agent setup
- Do not call wake repeatedly inside one uncompacted conversation; it bloats context and repeats stale summaries
- Never write directly to
core/ just because wake surfaced something important
- Wake is a hot-start summary, not a substitute for search on factual questions