By name: start-session description: Start a coding session by loading recent context, active tasks, and project constraints. Use at the beginning of each AI session to quickly get up to speed. allowed-tools: Bash, Read, Glob, Grep
/start-session — Session Startup Protocol
Trigger
/start-session [issue-number]
Input
- Optional: GitHub issue number to load specific context for
- Assumes current working directory is the project root
Execution Flow
Step 1: Sync spec (if submodule exists)
Check if spec/ directory exists and is a git submodule:
if [ -d spec/.git ] || [ -f spec/.git ]; then
git submodule update --remote spec/
fi
Step 2: Read recent context
Read the rolling context recovery file:
- Primary: Read
spec/Progress/LATEST.md(orMeta/Progress/LATEST.mdif no submodule) - Fallback: If LATEST.md doesn't exist, find the most recent file in
spec/Progress/(orMeta/Progress/) by filename sort - Read
spec/Todo.md(orMeta/Todo.md) for current task backlog
Step 3: Load issue context (if issue number provided)
If an issue number was provided as argument:
gh issue view {number} --json title,body,labels,milestone
Look for "Spec Context" or "Related Files" sections in the issue body. If found, read those spec files.
If no issue number but LATEST.md mentions active work, note the current focus area.
Step 4: Read constraints
Based on the task type inferred from LATEST.md or the issue:
- For code changes: Read
spec/Core/Regulation/CodeStyle.md+spec/Core/Regulation/Architecture.md - For test work: Read
spec/Core/Regulation/Testing.md - For git operations: Read
spec/Core/Regulation/GitWorkflow.md - Always: Scan
spec/Core/Regulation/Meta.mdfor the routing table
Step 5: Output summary
Print a formatted summary:
## Session Ready
**Last activity**: {summary from LATEST.md}
**Active work**: {from LATEST.md Active Work table}
**Current task**: {from issue or Todo.md}
**Constraints loaded**: {list of regulation files read}
**Blockers**: {from LATEST.md or "none"}
**Test count**: {last known test count from LATEST.md}
### Next steps
{Suggested actions based on context}
Constraints
- This skill is read-only — it does not modify any files
- If spec/ submodule pull fails (network), continue with local files
- Do not read more than 5 spec files to keep context window manageable
- Always check both
spec/andMeta/paths (support both submodule and non-submodule setups)