Context For Next Session

Generate a CONTEXT_FOR_NEXT_SESSION.md that enables a new Claude session to resume work with zero re-explanation.

Process

1. •

   Check for existing file — Read ./CONTEXT_FOR_NEXT_SESSION.md if it exists. If found:

   • •Preserve previous context under a ## Previous Session Context heading
   • •Add new content as the primary (top) section with updated timestamp
   • •Merge open items: check off completed ones, carry forward remaining
2. •

   Gather session state:

   • •Current git branch, uncommitted changes, recent commits
   • •Working directory and key file paths referenced in conversation
   • •Environment details (which env was targeted: local, QA, UAT, prod)
   • •Skills invoked during this session (scan conversation for /skill-name invocations)
3. •

   Extract from conversation:

   • •What we were working on and why
   • •Where we left off (specific point of progress)
   • •Decisions made that constrain future work
   • •Key files/modules/paths that are relevant
   • •Known issues, blockers, or gotchas discovered
   • •Suggested next actions in priority order
4. •

   Write CONTEXT_FOR_NEXT_SESSION.md in the current working directory.

5. •

   Confirm the file path.

Output Format

# Context For Next Session

**Date:** {YYYY-MM-DD}
**Branch:** {current git branch}
**Working Dir:** {cwd}
**Last Session:** {1-line summary of what was accomplished}

---

## What We're Working On

{2-3 sentences: the goal, why it matters, and current approach}

## Where We Left Off

{Specific stopping point — what was the last thing done/attempted}

## Key Context

### Relevant Files

- `{path/to/file}` — {why it matters}

### Environment State

- **Branch:** {branch name} — {clean/dirty, ahead/behind}
- **Target env:** {local/QA/UAT/prod}
- **Dependencies:** {anything unusual about current state}

### Decisions Made

- {Decision}: {rationale} — constrains {what}

### Gotchas & Discoveries

- {Thing that wasn't obvious but matters}

## Next Steps

1. {Highest priority next action}
2. {Second priority}
3. {Third priority}

## Skills To Load

{Auto-populated from skills invoked during this session. List each skill that was
actively used (not just mentioned). Include a brief note on what it was used for.
If no skills were invoked beyond session-kit lifecycle commands, omit this section.}

- `/beam-expert` — OTP process debugging
- `/use-db` — queried UAT eventstore

---

_Paste this document at the start of your next Claude Code session in this directory._

Rules

• •Optimized for machine consumption — this is for Claude to read, not just humans. Be precise about paths, branch names, and state.
• •Include the "why" not just the "what" — next session needs to understand intent, not just facts
• •Concrete over abstract — "check lib/my_app/accounts/projectors/user_projector.ex line 42" beats "look at the projector"
• •Skip sections with no content — don't include empty Gotchas or Decisions sections
• •Auto-detect skills — Scan the conversation for skill invocations (commands like /beam-expert, /use-db, etc.) and list them in Skills To Load. Exclude session-kit lifecycle skills (/park, /relay, /tldr, /pickup, /prompt-lab, /retro, /handoff, /rca, /index, /persist) — those are infrastructure, not domain context.
• •Skills recommendation — always include which skills were loaded/useful so the next session can load them immediately
• •Write to ./CONTEXT_FOR_NEXT_SESSION.md unless the user specifies a different path