| name | dev-execute |
| description | Execute specifications from configurable directory (defaults to ./.dev-docs/features/) in fresh sub-agent contexts. Two-phase workflow - Phase 1 implements and provides testing guidance, Phase 2 archives after user confirms testing complete. Supports single, parallel, and sequential execution. |
Execute feature specifications from configured directory as delegated sub-tasks with fresh context
Keeps main conversation clean while implementation happens in focused, isolated sub-agent contexts
Reads dev_docs_directory from root CLAUDE.md, defaults to ./.dev-docs/features/
Sub-agent implements feature specification
Provides suggested manual testing steps
Feature stays in configured directory (NOT archived yet)
User performs manual testing based on suggestions
User explicitly confirms "feature X is complete" or "archive feature X"
Feature archived to [configured_path]/completed/
Feature specifications have been created by dev-spec and are ready to implement
User wants to execute one or more saved feature specifications
Need to run multiple features in parallel or sequential order
User has tested a feature and wants to mark it complete and archive it
<tool_restrictions>
<allowed_tools>
Spawn sub-agents with subagent_type="general-purpose"
Read feature specification files, read CLAUDE.md for configuration
Find files, create directories, archive specifications
Search for feature files by pattern
</allowed_tools>
<critical_pattern>
<parallel_execution>ALL Task tool calls MUST be in a SINGLE message for parallel execution</parallel_execution>
</critical_pattern>
</tool_restrictions>
Read root CLAUDE.md for dev_docs_directory setting to determine spec directory (defaults to ./.dev-docs/features/ if not set)
NEVER archive features automatically after implementation - always wait for explicit user confirmation after manual testing
Sub-agents MUST provide suggested manual testing steps in their final report
All Task tool calls for parallel execution must be in a single message
For sequential execution, stop immediately on first failure
Each sub-agent gets fresh context with only the specification content
Verify configured directory exists before execution
Create [configured_path]/completed/ directory if it doesn't exist
When delegating to sub-agent, explicitly instruct them to provide manual testing steps
For parallel execution, report which features succeeded/failed separately
Archive format includes execution metadata (timestamp, status)
Clearly distinguish between "implemented" and "completed/tested" status
<directory_configuration>
Determine where feature specifications are stored - reads configuration set by dev-spec
Read root CLAUDE.md file (./CLAUDE.md)
<look_for>dev_docs_directory: [path]</look_for>
<if_found>Use [path]/features/ for all specification reads</if_found>
<if_not_found>Use default ./.dev-docs/features/</if_not_found>
Read specifications from [configured_path]/features/
Archive completed specifications to [configured_path]/completed/
<important_notes>
This skill only READS the configuration - dev-spec is responsible for setting it
Always check ROOT CLAUDE.md (./CLAUDE.md) - there may be multiple CLAUDE.md files in subdirectories
The path is project-specific - different projects can have different paths
If CLAUDE.md doesn't exist or doesn't have the setting, default to ./.dev-docs/features/
</important_notes>
</directory_configuration>
<execution_modes>
User requests to implement/execute one feature by number, name, or "latest"
Resolve feature file using matching strategy
Read complete specification content
Delegate to sub-agent via Task tool with instruction to provide testing steps
Wait for completion
Present implementation results and manual testing suggestions to user
Feature remains in ./.dev-docs/features/ awaiting testing confirmation
<no_automatic_archiving>Feature stays in features/ directory until user confirms completion</no_automatic_archiving>
Features are independent with no shared files or dependencies
User explicitly requests parallel execution
Read all requested feature specifications
Spawn ALL Task tools simultaneously in one message, each instructed to provide testing steps
Wait for all sub-agents to complete
Present consolidated implementation results and testing suggestions for each feature
Features remain in ./.dev-docs/features/ awaiting testing confirmation
Maximum efficiency but requires feature independence
All Task calls must be in ONE message
Features stay in features/ directory until user confirms completion
Features have dependencies or modify shared files
Order matters for correctness
Execution strategy not specified (default for safety)
Execute features one at a time in order, each instructed to provide testing steps
Wait for each completion before starting next
Stop immediately on first failure
Present implementation results and testing suggestions after each feature
Features remain in ./.dev-docs/features/ awaiting testing confirmation
Features stay in features/ directory until user confirms completion
User explicitly confirms feature is tested and ready to archive
User says "feature X is complete", "archive feature X", "mark feature X done"
Resolve feature file in ./.dev-docs/features/
Verify feature file exists (not already archived)
Add completion metadata (timestamp, status)
Move to ./.dev-docs/features/completed/
Confirm archival to user
feature X is complete
archive feature X
mark feature X done
complete feature X
<manual_testing_guidance>
All sub-agents MUST include suggested manual testing steps in their final report
<sub_agent_instruction>
When delegating to sub-agent, include this instruction in the prompt:
"After implementing the feature, provide a section called 'Suggested Manual Testing Steps' with specific, actionable steps the user should perform to verify the feature works correctly."
</sub_agent_instruction>
<testing_step_quality>
Test user login with valid credentials at http://localhost:3000/login
Verify API endpoint returns 200 status: curl http://localhost:8080/api/users
Check database has new table: SELECT * FROM user_sessions;
Test the feature
Make sure it works
</testing_step_quality>
After sub-agent completes, present testing steps clearly to user and remind them to confirm completion after testing
</manual_testing_guidance>
<feature_resolution>
Match zero-padded numbers (5 matches 005-.md)
[configured_path]/features/
Find files containing name string in filename
auth matches 005-user-authentication.md
ls -t [configured_path]/features/.md | head -1
<conflict_resolution>
<one_match>Use that file</one_match>
<multiple_matches>List options and ask user to choose</multiple_matches>
<no_matches>Report error and list available features</no_matches>
</conflict_resolution>
configured_path comes from dev_docs_directory in root CLAUDE.md, defaults to ./.dev-docs
</feature_resolution>
[configured_path]/completed/
[number]-[name].md
executed: [ISO timestamp of implementation]
completed: [ISO timestamp of archival]
status: completed
Only after user explicitly confirms manual testing is complete
User must say "feature X is complete" or similar confirmation phrase
configured_path comes from dev_docs_directory in root CLAUDE.md, defaults to ./.dev-docs
<error_handling>
Report failure with error details, do not archive, feature stays in features/
Report which succeeded/failed, do not archive any, all stay in features/
Stop immediately, do not execute remaining features, failed feature stays in features/
List available features in [configured_path]/features/
Ask user to clarify which feature to execute
List all matching feature files
Ask user to specify by full number or more specific name
User tries to complete/archive a feature that's already in completed/
Inform user the feature is already archived
Show location in [configured_path]/completed/
</error_handling>
Execute feature 005
Find [configured_path]/features/005-user-authentication.md
Read specification
Spawn Task tool with specification + instruction to provide testing steps
Wait for sub-agent completion
Present implementation results + suggested manual testing steps
Remind user: "Feature 005 has been implemented. Please test using the steps above, then confirm completion when ready."
Feature remains in [configured_path]/features/005-user-authentication.md
Feature 005 is complete
Find [configured_path]/features/005-user-authentication.md
Add completion metadata (executed timestamp, completed timestamp, status)
Move to [configured_path]/completed/005-user-authentication.md
Confirm: "Feature 005 archived to [configured_path]/completed/"
Execute features 005, 006, and 007 in parallel
Read all three feature specifications
Spawn three Task tools in single message, each instructed to provide testing steps
Wait for all completions
Present consolidated results with testing steps for each feature
Remind user: "All three features implemented. Please test each one, then confirm completion individually or together."
All three features remain in [configured_path]/features/
All Task calls must be in ONE message
Features 005, 006, and 007 are all complete
Archive 005-user-authentication.md to completed/
Archive 006-api-endpoints.md to completed/
Archive 007-ui-components.md to completed/
Confirm: "All three features archived to [configured_path]/completed/"
Execute features 005, 006, 007 sequentially
Feature 006 fails during implementation
Execute feature 005 successfully, provide testing steps
005 remains in [configured_path]/features/ awaiting testing
Execute feature 006, encounters error
Stop execution, do not run feature 007
Report failure at feature 006, no archival, both 005 and 006 stay in features/
Execute feature 005
Implementation complete. Suggested manual testing steps:
1. Start the application: npm run dev
2. Navigate to http://localhost:3000/login
3. Enter test credentials: user@example.com / password123
4. Verify successful login and redirect to dashboard
Feature 005 has been implemented. Please test using the steps above, then let me know when it's complete.</agent>
<user>*user tests the feature*</user>
<user>Feature 005 is complete</user>
<agent>Feature 005 has been archived to [configured_path]/completed/005-user-authentication.md</agent>
</conversation>
<related_skills>
Orchestrator for overall feature development workflow
Creates feature specifications that this skill executes
For simpler tasks rather than complex features
</related_skills>