resume

Resume Research

Continue a paused or interrupted research-agent run. If the user names a branch (e.g. "resume the induction-proof branch" or /research-agent:resume induction-proof), verify .research/branches/<name>/ exists and add <branch>name</branch> to the launch prompt; all checks below then apply to the branch subtree's files.

Step 1: Verify .research/ exists

Use the Read tool on .research/state.md. If state.md is missing but .research/goal.md exists (a run interrupted before its first REFLECT), still resume: the researcher reorients from goal.md and log.md and rebuilds state.md. Only when .research/ itself is absent should you tell the user there is no research to resume and suggest /research-agent:research.

Step 2: Quick state preview

Use Read tool to load .research/state.md and the last 5 entries from .research/log.md. Show the user a one-paragraph preview of what the researcher will pick up:

Resuming research on "{goal}". Current focus: {from state.md}. Last cycle was {N} on {date}: {modality} -> {result}. Outstanding sub-problems: {count open}. Untested hypotheses: {count untested}.

Step 3: Launch the researcher with resume framing

Spawn the researcher agent (Agent tool, subagent_type research-agent:researcher) with this prompt:

<mode>resume</mode>

<working_directory>
{absolute path to current working directory}
</working_directory>

Resume your research from disk. Do NOT recreate .research/ or
overwrite goal.md.

Step 1: Read .research/state.md in full to reload current beliefs,
sub-problem statuses, hypothesis statuses, and current focus.

Step 2: Read the last 15 entries of .research/log.md to recall recent
activity, including any failed attempts and pivots.

Step 3: Continue your research loop from the documented "Current
focus" in state.md. Run cycles as normal: DECOMPOSE if needed,
HYPOTHESIZE, ATTEMPT, EVALUATE, REFLECT. Append to log.md after
every cycle.