agentsclimarketplace

Session handoff

Skill OthmanAdi/MDDesign/.opencode/skills/session-handoff

Makes sessions durable across `/clear`, machine reboots, and IDE switches. SessionStart catchup pulls scratch + last MemPalace recall + planning files into context. Stop promotion writes a handoff note and runs memory-keeper. Manual `/mddesign:handoff` re-runs either at any time.From its SKILL.md

Install
npx -y skills add OthmanAdi/MDDesign --skill session-handoff

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 13 stars13 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

4.8 KB, as published. Nobody here has run it

Session Handoff

Bridges sessions. The user's session ends; the next session picks up exactly where the last one left off.

SessionStart catchup

Triggered automatically by the SessionStart hook. Manually invokable via /mddesign:handoff catchup.

Step 1: Read planning files

  • Read task_plan.md if it exists. Note the active phase id and its bullets.
  • Read findings.md if it exists. Note any ## Design Context block.
  • Read progress.md if it exists. Note the most recent ### Phase <id> Result and the most recent ### Handoff <timestamp> entry.

Step 2: Read scratch

  • Glob .agents/memory/scratch/**/*.md.
  • For each scratch file, read the last ISO-timestamped entry.

Step 3: Recall last WHY

  • Invoke /memory-router with a query like "recent decisions for project <CWD basename>".
  • The router dispatches to MemPalace; returns the most recent WHY entries for this project.

Step 4: Synthesize the preamble

Build a one-paragraph (4-6 sentence) "where we left off" that names:

  • The active phase and its current status
  • The most recent decision (from WHY)
  • Any open questions from the last ### Phase Result
  • Any in-flight scratch entries

Print this preamble to the user as your first message.

Step 5: Surface nudges

  • If the project has UI files (*.tsx|*.vue|*.svelte) but no DESIGN.md, print: "No DESIGN.md found. Run /mddesign:harvest to generate one."
  • If task_plan.md does not exist, print: "No task_plan.md. Use planning-with-files to create one before starting work."
  • If code-memory-router slash command is not registered, print: "code-memory-router not detected. Memory operations will fall back to scratch only."

Stop promotion

Triggered automatically by the Stop hook. Manually invokable via /mddesign:handoff promote.

Step 1: Dispatch memory-keeper

Call the Agent tool with subagent_type: memory-keeper. Pass:

  • A pointer to progress.md (last 50 lines)
  • A pointer to findings.md
  • A pointer to .agents/memory/scratch/

Memory-keeper returns a list of promotion candidates with tier classifications.

Step 2: HITL on each promotion

For each WHY candidate:

  • Print the candidate text and the proposed MemPalace drawer path.
  • Wait for the PreToolUse hook (named checkpoint: before_mempalace_write) to require approval.
  • On approval, the memory-layer skill performs the actual /memory-router invocation.
  • On rejection, leave the entry in scratch.

Step 3: Write the handoff note

Append to progress.md:

### Handoff <ISO timestamp>

**Active phase at session end:** <id>
**Status:** <status>
**Promoted to WHY:** <count> entries
**Open questions:**
- <list>

**Resume next session by:**
- <one-line action>

This block is what SessionStart catchup reads first next time.

Manual handoff

SubcommandWhat it does
/mddesign:handoff (no arg)Runs catchup.
/mddesign:handoff catchupRuns catchup.
/mddesign:handoff promoteRuns Stop promotion. Useful before a planned /clear.

Core rules

  • Never touch task_plan.md directly. Read only.
  • Never write to MemPalace without HITL.
  • Catchup is read-only. Only the promotion step writes (and only under HITL).
  • The handoff note in progress.md always lives under ### Handoff <timestamp> for easy grep.

Example

User ends Tuesday session mid-Phase 3. Stop fires. Memory-keeper proposes 2 WHY candidates ("decided X for reason Y" and "rejected approach Z"). User approves both. Plugin writes:

### Handoff 2026-04-23T18:42:11Z

**Active phase at session end:** phase-3-build-cta
**Status:** in_progress, button component scaffolded but a11y not yet checked
**Promoted to WHY:** 2 entries
**Open questions:**
- Should the focus ring use colors.primary or colors.accent?

**Resume next session by:**
- Read findings.md "## Design Context", then run /mddesign:critique to check the button against DESIGN.md tokens.

Wednesday morning, user opens a new session. /clear happened or the laptop rebooted. SessionStart fires. Catchup synthesises:

Last session ended in Phase 3 (build-cta), button scaffolded but a11y not yet checked. Decisions promoted: chose primary color for CTA based on contrast vs accent. Open question: focus ring color (primary vs accent). Suggested next step: read findings.md "## Design Context" then /mddesign:critique.

User reads, says "ok continue". The agent already has every piece of context it needs.

Keep looking

Skills are one crate of 627,801. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.