Setup and quick reference
Project 42 Daily AI Work Quick Reference
Use a compact orient, plan, execute, verify, and recover loop for everyday AI-assisted work, with deeper Project 42 guides for specialized decisions.
- Format
- Reference
- Level
- Beginner
- Audience
- Learner, Practitioner, Developer, Operator
- Owner
- Project42 Editorial
- Review cadence
- Every 60 days
- Prerequisites
- An authorized task and target workspace; A way to review changes and observable results; A known recovery point
Orient and plan
Confirm the requested outcome, target, owner, repository or system instructions, current state, protected data, allowed actions, and stop conditions. Separate facts from assumptions and identify the smallest evidence that will prove success.
Use AI Coding Workspace Bootstrap for first-time setup, Agent Configuration Layering Reference for conflicting settings, and the relevant coding-tool operating card when choosing permissions and interaction style.
Execute in small observable steps
Inspect before editing, make the smallest coherent change, and keep unrelated user work intact. Review every new dependency, file, command, network call, permission expansion, and external write. Capture useful outputs and stable identifiers without secrets or unnecessary personal data.
Use Commands, Hooks, and Automation Guide before turning a repeated step into executable automation. Use the MCP trust and tool-contract guides before attaching new tools or servers.
Task: [REQUESTED OUTCOME]
Scope: [TARGETS, ALLOWED PATHS, AND EXCLUSIONS]
Permissions: [READ, WRITE, COMMAND, NETWORK, TOOLS, AND EXTERNAL ACTIONS]
Known state: [BRANCH, VERSION, STATUS, OR RESOURCE ID]
Evidence needed: [TEST, DIFF, SOURCE, SCREEN, OR POSTCONDITION]
Stop conditions: [SCOPE CHANGE, SECRET, UNSAFE ACTION, UNKNOWN WRITE, OR FAILED GATE]
Verification: [RUN CHECKS, REVIEW COMPLETE CHANGE, TEST USER BOUNDARY]
Recovery: [STOP, PRESERVE EVIDENCE, RECONCILE, RESTORE, AND RETEST]Expected result and verification
Expected evidence includes the original outcome, actual scope, material actions, reviewed changes, source or runtime evidence, tests, user-boundary result, unresolved risk, and recovery status. A successful command is not enough when the user-facing or external postcondition is still unverified.
When the result is wrong or uncertain, stop before another consequential action. Preserve redacted evidence, determine whether any external write occurred, restore the last verified state or apply an approved compensation, rerun the smallest deterministic check, and update the relevant deeper guide or regression test.