skills/swe-wm-update/SKILL.md
Comprehensive Working Memory update with per-step checklists using swe-wm MCP tools
npx skillsauth add earthmanweb/serena-workflow-engine swe-wm-updateInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
Single source of truth for WM updates. Uses swe-wm MCP tools for targeted section
updates — never overwrites daemon-managed fields.
Every WF_* transition MUST use this skill instead of manual updates.
NEVER use write_memory or edit_memory to update WM files. These overwrite the
entire file and risk clobbering daemon-managed fields.
ALWAYS use swe-wm MCP tools:
| Tool | Purpose |
| ------------------------- | --------------------------------------------- |
| swe_wm_read | Read WM state + content |
| swe_wm_update_section | Update a specific section (agent-owned only) |
| swe_wm_update_status | Update **[STATUS]**: tag |
Daemon-managed fields (updated automatically by Python hooks — DO NOT touch):
**Current State**: — updated on WF_* reads**Previous State**: — updated automatically### Transitions — appended automatically**Edit Count Since Checkpoint**: — incremented on file edits**Last Updated**: — timestamped automaticallymcp__swe-wm__swe_wm_read(session_id="{session_id}")
Note the current state and section contents for your update decisions.
Find the checklist matching your --from argument below. Complete ALL items.
Current Task section with task + contextFeature(s) sectionProgress with initial itemsAffected Features with Primary / SecondaryFiles with key file paths from feature memoriesProgress with classification and feature loading steps completedFiles with files examinedProgress with research outcomesNotes with findingsFiles with files from previous workFiles with files to modify/createProgress with review resultsNotes with design decisionsProgress with orchestration stepsNotes with swarm configFiles with files modified / created with descriptionsProgress with implementation stepsFiles with files modified with change descriptionsProgress comprehensivelyFiles — complete list of every file touched in sessionVERIFY_COMPLETE or COMPLETEDCOMPLETEDContext with summary of all work doneNotes with memories updated during sessionProgress items checked offProgress with clarification outcomeUse targeted section updates. One call per section that changed.
# Update task status
mcp__swe-wm__swe_wm_update_status(session_id="{session_id}", status="IN_PROGRESS")
# Update specific sections (replace mode)
mcp__swe-wm__swe_wm_update_section(
session_id="{session_id}",
section="Progress",
content="- [x] Step 1 done\n- [ ] Step 2 pending"
)
# Append to a section (e.g., adding a new file entry)
mcp__swe-wm__swe_wm_update_section(
session_id="{session_id}",
section="Notes",
content="- New finding: ...",
append=true
)
Agent-owned sections (safe to update):
Current Task, Progress, Files, Notes, Requirements,
Implementation Notes, Previous Task, Task Context,
Affected Features, Context, Feature(s)
Protected sections (tool will reject these):
Workflow Context, Transitions
After writing, optionally re-read to confirm:
mcp__swe-wm__swe_wm_read(session_id="{session_id}")
Verify:
Output: 📋 Updated Working Memory: WM_{session_id}
⚠️ CRITICAL: DO NOT STOP HERE. This skill is a utility — you MUST continue.
IMMEDIATELY resume the workflow step you were on before invoking this skill.
This is a utility skill — no state change occurs. Your calling step's instructions
told you to invoke /swe-wm-update as a sub-step, NOT as a stopping point.
After outputting the confirmation line above:
If you were invoked from a WF_* step's "MANDATORY NEXT STEP" table, proceed to the transition listed there. If you were invoked mid-step, continue where you left off in that step's instructions.
testing
Verify implementation against requirements and standards
development
Code exploration and research without making changes
development
Test-driven debugging for failing tests or bugs
tools
Review architecture compliance before execution