Deepdone Verify

DeepDone Verify

Purpose

Make verification deterministic and scoped.

Verification Order

Start narrow. Widen only when impact demands it.

Typical order:

1. focused unit or smoke check nearest change
2. package or app lint
3. package or app typecheck
4. affected test suite
5. broader integration or full gate if shared code changed

If repo exposes shared wrappers, prefer just . If not, use repo's real package-level commands.

Command Discovery

Before guessing commands, inspect repo affordances in this order:

1. justfile or .justfile: prefer named project wrappers.
2. Makefile: use focused targets like test, lint, check, or package-specific targets.
3. package.json: read scripts; prefer focused package scripts before broad test.
4. pyproject.toml: inspect tool config and project layout; prefer uv run, pytest, ruff, mypy, or repo wrappers already documented.
5. Cargo.toml: prefer cargo test, cargo check, or package-specific variants.
6. go.mod: prefer go test ./... only when package-level test is not obvious.
7. Existing CI config or README commands if no local wrapper is found.

Record discovery when it affects command choice:

- command: discovery
  result: pass
  notes: found `justfile`; using `just test-auth` before broader checks

If discovery finds no runnable command, log a blocked entry with exact reason.

Mandatory Notes

Record exact command and result. If something could not run, state exact blocker. Do not claim pass based on intent.

Ledger Writeback

Append lines under ## Verification Log such as:

- command: `just test api-auth`
  result: pass
  notes: focused auth test passed
- command: `uv run pytest tests/test_auth.py`
  result: fail
  notes: login fixture broken after schema change
- command: `just lint`
  result: blocked
  notes: no `justfile` in repo

result: is required. Use only pass, fail, or blocked.

Risk Triggers

Broaden verification when change touches:

• shared libs
• auth or permissions
• schema or migrations
• external APIs
• config or deployment-sensitive paths
• user-visible flows

Workflow

1. Discover command wrappers and package-level checks.
2. Identify smallest convincing check.
3. Run it.
4. Decide whether broader checks are needed.
5. Run broader checks only where justified.
6. Log every command and outcome with result:.
7. State residual risk if coverage still incomplete.

Output

Return:

• checks run
• pass, fail, or blocked status
• residual risk

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: lakesoftai
• Source: lakesoftai/deepdone-agent-skills
• License: Apache-2.0
• Homepage: https://deepdone.ai

Install and usage instructions live in the source repository linked above.