handoff

name	handoff
description	Create a comprehensive continuation plan so a fresh Hermes session can pick up exactly where this one left off — with full operational detail, not summaries.
queries	["hand off this work","create a handoff","continue in a new session","save progress for next session","context window getting long","pick this up later","pass this to a new context","write a continuation plan"]

/handoff — Create a Comprehensive Continuation Plan (Hermes)

Capture the full operational state of the current work so a completely fresh Hermes session can resume without losing any context. The next session has zero memory of this one — every detail that matters must be in the document.

Budget: Spare no necessary detail. A handoff that's too short forces the next session to re-discover context, wasting hours. A handoff that's comprehensive saves the user from repeating themselves. Err on the side of completeness.

When to Use

• Context window is getting long and performance is degrading
• Switching to a different machine or environment
• Pausing work to resume later
• Handing off to a colleague or a different agent

Process

1. Summarize the objective

Write a clear statement of what the user is trying to accomplish, with enough context that someone unfamiliar could understand the goal. Include:

• The high-level goal
• Why it matters (business/technical motivation)
• How this session fits into the larger picture

2. Document what's been done — IN DETAIL

For each PR merged or significant piece of work:

• PR number, title, and what it actually changed (not just the title — describe the mechanics)
• Files modified with the specific changes and why
• The problem it solved with the root cause, not just the symptom
• Verification: what tests were run, what CI status was, what review found
• Operational impact: what needs to happen after merge (deploy commands, service restarts, etc.)

For decisions made during the session:

• What was decided and the specific rationale
• What alternatives were considered and why they were rejected
• Supporting data (benchmark numbers, cost analysis, error messages)

For investigations/research:

• What was checked and the specific findings
• File paths and line numbers for key code examined
• Conclusions with evidence, not just assertions

3. Document current system state

Capture the actual state of every system touched, with verification commands. Include the actual output, not just the commands. The next session should be able to verify state without re-running.

# Git state
git branch --show-current
git log --oneline -15
git status
git worktree list

# Open PRs
gh pr list --author @me --state open

# Hermes session identity (for resume / pairing)
echo "$HERMES_HOME"
ls "$HERMES_HOME/sessions/" | tail -5

# Service health (if applicable)
# Include actual JSON responses, not just "it's healthy"

4. Document what's left — with full context per item

For each remaining item, provide:

• What needs to happen — concrete, actionable steps (not "fix the thing")
• Where in the codebase — specific file paths, function names, line numbers
• Why it matters — what breaks or is blocked without it
• Blocked by — dependencies on other items
• Suggested approach — if one was discussed or is obvious from the session
• Known pitfalls — things the next session should watch out for
• Verification — how to confirm the item is done correctly

Order items by priority or logical sequence.

5. Document failed approaches and dead ends

For each approach that was tried and abandoned:

• What was tried and the specific error or problem
• Why it failed — root cause, not just "it didn't work"
• What was learned that should prevent the next session from retrying

This section prevents the most expensive waste: a fresh session re-discovering the same dead ends.

6. Capture operational knowledge and gotchas

Everything that would be lost with the context window:

• Environment quirks with exact workaround commands
• Non-obvious dependencies between components
• Things that look like bugs but aren't (and why)
• Service state that affects behavior (what's deployed where, what version)
• Credentials/access patterns (not the secrets — the patterns for using them)
• User preferences observed during the session that aren't yet saved as rules
• Hermes-specific surface area touched: memory.provider config key, which sandbox profile, which agent_identity, whether agent_context was anything other than primary

7. Write the handoff file

Write to: <cwd>/.hermes/handoffs/<YYYYMMDD>-<kebab-case-title>.md

Use this structure:

# Handoff: <descriptive title>

**Date:** <ISO date>
**Branch:** <current branch>
**Working directory:** <cwd>
**Hermes session ID:** <session-id from /resume target, if known>

## Objective

<What we're trying to accomplish and why. 2-3 paragraphs of context.>

## Session Summary

<One paragraph narrative of what happened this session — the arc from start to finish.>

## Completed Work

### <PR    >

 #NNN — 
 
 
 
 
 
 
 





| Decision | Rationale | Alternatives Rejected |
|----------|-----------|----------------------|
|  |  |  |








\


\


\


 : , 
 : , 





 
 
 
 
 
 
 






 
 
 



 
 



 Read this file: 
 
 
 If continuing the same session:  (the ID is at the top of this file)

8. Confirm with the user

Show the handoff document and ask:

• Is anything missing?
• Should any items be reprioritized?
• Are there preferences or context to add?

Update the file with any corrections.

9. Tell the user how to take over

To continue, start a new Hermes session and either:

  hermes --resume <session-id>          # resume the same Hermes session

or read the handoff into a fresh session:

  cat .hermes/handoffs/<YYYYMMDD>-<title>.md

The handoff file itself is also indexed by memex on the next prefetch — if the new session lands on a relevant prompt, memex will surface the handoff semantically.

Guidelines

• Spare no necessary detail. The next session has zero context. Everything that matters must be written down. A long handoff that prevents hours of re-discovery is worth it.
• Be specific, not generic. File paths, branch names, commit hashes, PR URLs, exact error messages, benchmark values, cost figures — anything the next session would need to look up, include directly.
• Explain the why. Decisions without rationale will be re-evaluated from scratch.
• Include failed approaches. The most expensive waste is a fresh session re-discovering the same dead ends.
• Include actual command output. Don't say "check the service health" — run the command and include the output.
• Include operational commands. Every "after merge, do X" should have the exact command, not a description.
• Include the data. Tables of results, cost comparisons, benchmark numbers — these belong in the handoff, not just references to them.
• Capture the Hermes surface. Note HERMES_HOME value, active memory.provider, the session ID for --resume, and any non-default agent_context so the next session can re-attach to the right scope.
• Keep it scannable. Use headers, bullet points, tables, and checkboxes — but don't sacrifice detail for brevity.

$ARGUMENTS