| name | Engagement Context |
| description | Automatically loads Wire engagement and release context at the start of a conversation. Fires when a .wire/ directory is present and context has not yet been established in the current session. |
| triggers | ["Any message in a repository containing a .wire/ directory where engagement context has not yet been loaded","User begins discussing an engagement, release, or artifact without prior context in the conversation","User asks what to work on next, what the current status is, or what release is active"] |
Engagement Context Skill
On Activation
Before proceeding with any work, append a one-line entry to .wire/execution_log.md (at engagement root, or within the active release if no engagement-root log exists):
| YYYY-MM-DD HH:MM | skill | engagement-context | activated | Context loaded for new conversation |
If .wire/execution_log.md does not exist, create it with the standard header first.
When This Skill Activates
This skill fires automatically at the start of a conversation in any repository that contains a .wire/ directory, when the agent has not yet established engagement context. It is the replacement for the deprecated /wire:session:start command — context loading is now implicit, not explicit.
Do NOT fire this skill if:
- You have already read and summarised the engagement context earlier in this conversation
- The user's message is clearly not Wire-related (e.g. they are asking about something unrelated to the engagement)
Context Loading Procedure
Step 1: Locate the Active Release
Check .wire/releases/*/status.md (two-tier layout) or .wire/*/status.md (legacy layout). Sort by last modified; use the most recently modified.
If multiple releases exist, note them all. Use the most recently modified as the active release for the context summary.
Step 2: Read Engagement Context
Read .wire/engagement/context.md if present. Extract:
- Client name and engagement overview
- Business objectives
- Key stakeholders
- Current engagement state (which releases exist, which are active/completed)
Step 3: Read Release State
Read the active release status.md. Extract:
- Release name, type, and current phase
- Artifact completion state (what is done, in progress, blocked)
- Last 3 rows of session history (what was done recently and what the suggested next focus is)
- Any blockers
Step 4: Surface Context Summary
Output a brief, scannable context block before responding to the user's request:
## Wire Engagement Context
**Client**: [client_name] | **Release**: [release_name] ([release_type])
**Phase**: [current_phase] | **Last worked**: [date of last session history row, or "no prior sessions"]
**Status**: [X artifacts complete, Y in progress] — [brief 1-line state of play]
**Suggested next**: [next focus from last session, if any]
Keep this to 4–6 lines. Do not dump the entire status file. If there is no prior session history, say so in one line.
Step 5: Proceed With the User's Request
After outputting the context summary, immediately proceed with what the user asked for. Do not ask for permission to continue or offer a menu of options unless the user's request is ambiguous.
If the user asked a specific question or gave a specific instruction, answer it — do not replace their intent with a session plan. Reserve structured planning for /wire:plan.
Edge Cases
No .wire/ directory
Do not activate. Proceed normally without Wire context.
First session (no prior history)
Output the context summary with "No prior sessions" for the last-worked field and suggest the first incomplete artifact as the natural next step.
Multiple active releases
List all releases in one line: "3 releases: 01-discovery (complete), 02-data-foundation (active), 03-dashboards (planned)." Focus the state detail on the most recently modified.
engagement/context.md missing
Load release state only. Note "No engagement context file found" in one line — non-blocking.
What This Skill Does Not Do
- It does not enter Plan Mode or propose a session plan (use
/wire:plan for that)
- It does not block the user's request while loading context
- It does not ask the user questions before proceeding
- It does not replace the content of what the user asked — it prepends context to it
