Skip to content
Back to skills

End

ASecurity

End session — handoff + decisions + CLAUDE.md.

  • 18 stars
  • 0 votes
  • 0 copies
  • 1 view
  • Added October 6, 2026
ai-agentsgogit

Works with

  • cli
  • mcp

Security analysis

A100/100

Scanned October 6, 2026

npx -y skills add Kibertum/tausik-core --skill end --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of End?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for End
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/kibertum-end/badge)](https://www.skillsdirectory.com/skills/kibertum-end)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: end
description: "End session — handoff + decisions + CLAUDE.md."
effort: fast
context: inline
---

# /end — Session End (SENAR-aligned)

Archives session, updates CLAUDE.md for next session.
## Algorithm

### 1. SENAR Metrics Dashboard

Use `tausik_metrics` MCP tool to get metrics.

Display prominently:
- **Throughput**: tasks/session (is it improving?)
- **FPSR**: first-pass success rate
- **DER**: defect escape rate
- **Targets**: quote the report's `target … (basis: …)` lines as printed. Do not
  restate a number from memory: a target without its basis is not a target
  (SENAR 1.5 §9.4(c)). A `CROSSED` line is escalated to the owner, not moved.
- **Knowledge Capture Rate**: entries/task

### 2. Save Context (delegates to /checkpoint logic)

Run in parallel (MCP-first):
- `tausik_session_current` — get session state
- `tausik_task_list` with `status=active` — find in-progress work
- `tausik_status` — overall project state

Call `tausik_session_handoff` (CLI: `tausik session handoff`). The handoff is
**generated from the journal** — completed tasks, active tasks with their last
log line, verify receipts, decisions, memory, dead ends, open exploration. Do
not re-type what the records already say. Pass only your judgement, if any:

```json
{"next_steps": ["Continue task-slug-2"], "warnings": ["MCP server needs restart"],
 "in_progress": [{"slug": "task-slug-2", "state": "step 3 of 5"}]}
```

These land on top, marked in `authored_fields`.

**Note:** `session end` generates the handoff itself when none was written in the session, so a session never ends without one (SENAR §7.3).

### 3. Record Decisions

If any architectural or design decisions were made during the session, use `tausik_decide` with `decision="We chose X over Y"`, `rationale="Because Z"`, optionally `task_slug="{slug}"`.

### 4. Save Patterns and Dead Ends

If any reusable patterns, gotchas, or dead ends were discovered:
- Use `tausik_memory_add` with `type="pattern"` (or gotcha/convention/context/dead_end), `title="Short title"`, `content="Detailed description"`, optionally `tags=["tag"]`, `task_slug="{slug}"`
- **SENAR Rule 9.4:** If any approach failed and was NOT already documented, use `tausik_dead_end` MCP tool

Only save project-specific patterns — not framework instructions.

### 5. End Session

Use `tausik_session_end` with `summary="Completed X, fixed Y, planned Z"`.

### 6. Update CLAUDE.md

Use `tausik_update_claudemd` MCP tool to refresh the dynamic section.

### 7. Git Commit Prompt

Ask the user: "Commit changes? (y/n)"
- If yes — read `harness/skills/commit/SKILL.md` and follow its full algorithm inline (do NOT launch as subagent — commit needs user confirmation which requires inline context)
- If no — done
- Never force-push or commit without explicit approval

## Gotchas

- **Authored next steps go in before `session end`** — the generated part is written by `session end` anyway; your judgement is not.
- **Decisions should be recorded before ending** — they're linked to the session.
- **Dead ends must be documented** (SENAR Rule 9.4) — check if any failed approaches weren't recorded.
- **Don't save framework instructions to memory** — only project-specific patterns.

**Final message:** "Session closed. Start a new session with `/start` when ready."

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.