| name | dev-explore |
| description | Internal read-only reconnaissance for dev planning. |
| user-invocable | false |
| disable-model-invocation | true |
Announce: "I'm using dev-explore to map the relevant code and test path."
!bun ${CLAUDE_SKILL_DIR}/../../scripts/load-constraints.ts dev-explore
Read-only reconnaissance
The opening dev clarification sentinel must authorize this session before any discovery. Do not
write planning artifacts. Do not design or implement.
Use targeted, parallel read-only exploration when it helps. Trace the actual entry point, data
flow, integration boundaries, relevant conventions, similar implementations, and the test
infrastructure. Read every key file selected from exploration; summaries are not a substitute.
Return these findings directly in the conversation:
- Key files — 5–15 repo-relative
path:line entries and their purpose.
- Current architecture and conventions — dependencies and interfaces that constrain design.
- Real-test contract — framework, runnable command, production protocol/transport, user
action sequence, visible result, and test files/patterns to extend.
- Risks and ambiguities — decisions that code cannot answer.
Iron law: a test must do what the user does. Testing an alternate HTTP path when production
uses WebSocket, or inspecting source/logs instead of exercising runtime behavior, is not evidence.
If no usable automated harness exists, identify setup as the first planned task and obtain the
user's decision; never treat absence of tests as a waiver.
When reconnaissance exposes uncertainties, immediately load skills/dev-clarify/SKILL.md. Do
not write EXPLORATION.md, update a spec, or create a handoff/ledger.
