What makes an agent's work inspectable within the SDLC?
Why A is the answer
Standard artifacts such as commits, pull requests and logs
Artifacts in normal development channels can be reviewed, audited and reverted by the team.
In detail
- Commits, pull requests and logs are standard artifacts.
- People can inspect them with normal tools.
- Use the same artifacts that people produce.
Use case
An illustrative scenario, not a real incident.
A reviewer wants to know what an agent did. Its commits, pull request and logs show this with normal tools, rather than a private summary.
See it in different ways
sequenceDiagram participant D as Developer participant A as Cloud agent participant G as Repository participant R as Reviewer D->>A: Assign a task A->>G: Push to a copilot branch A->>G: Open a draft pull request R->>G: Review, then approve and merge
sequenceDiagram participant D as Developer participant A as Cloud agent participant P as Another reviewer D->>A: Requests a pull request A->>P: Draft pull request P->>P: Approve and run workflows P->>A: Review and merge
mindmap
root(("Agent work"))
n1["Commits"]
n2["What changed"]
n3["Pull requests"]
n4["Why and for review"]
n5["Logs"]
n6["What ran"]Key point: Standard artifacts make agent work inspectable.
Why the other options are wrong
- B. A longer system prompt describing how careful the agent should be
- A prompt describes intent, not what happened.
- C. A private chat summary that only the agent can read back later
- A private summary cannot be inspected.
- D. A final message saying that everything went according to plan
- A final message is a claim, not evidence.