Re-reading files already in context
Agents often read the same file range again even though it is still in the current context. ClevAgent uses three levels of response: it interrupts a large, high-confidence duplicate before the extra read; it gives guidance after the turn when the overlap is useful but not identical; and it quietly records small or uncertain overlaps. A changed file or a non-overlapping range is never treated as reusable old context.
For a large, high-confidence repeat, ClevAgent stops the duplicate read before it runs and directs the agent back to existing context. Smaller overlaps get guidance after the turn, while uncertain cases stay in the background.
Forgetting its own writes
Sometimes an agent reads back a file it just wrote or edited itself. It already produced that content, so the readback usually means it lost track of its own work. This Rule coaches the agent after the read rather than blocking it in advance, and it backs off if another process may have changed the file.
ClevAgent lets the read finish, then reminds the agent to use the content it just wrote. It stays quiet when another process may have changed the file.
Writing files in the wrong place
Reusable notes, reports, and other work files can get lost when they land in a temporary or unrelated folder. This Rule checks newly created work files and suggests a more findable home based on the project, filename, and active workspace. It stays quiet for source code, tests, configs, existing files, and files already in a sensible place.
ClevAgent suggests a more findable location for reusable notes and reports. Temporary work, source code, tests, configs, and sensible existing paths are left alone, and nothing moves automatically.
Piling everything into one file
Working files that grow past 1,200 lines can slow later search, edits, and context loads. ClevAgent suggests safe cleanup only after it can confirm the size. For code and other structured files, it asks the agent to review and report back without changing the file. It stays quiet when you explicitly asked for one long file and skips formats where line count is not useful, including generated assets, lock files, logs, and tabular data.
When reusable text grows past 1,200 lines, ClevAgent asks the agent to review whether splitting, indexing, or trimming would help. Code and other structured files get a review-only prompt, not an automatic edit.
Letting Claude project memory grow
Claude project memory can grow until recall gets noisy. Every 10 user turns, ClevAgent checks the active project's memory size and whether MEMORY.md has crossed Claude's 200-line or 25 KB startup window. It stays silent for Codex and never changes memory unless you send the suggestion.
Every 10 user turns, ClevAgent checks the active Claude project’s startup memory window. If it has grown past the limit, ClevAgent offers a project-scoped cleanup for you to send. It never changes memory on its own.
Crowded document folders
Document-heavy folders become harder for agents to scan and easier for useful work to lose. This Rule notices measured signs of crowding after a new file is added and offers to help organize the folder. It does not move or delete anything on its own; you decide whether to send the suggestion.
After a new file makes a document folder look crowded, ClevAgent offers an organization review. It does not move or delete anything on its own.
Code changed without a successful check
An agent can finish after changing code without confirming that the updated code works. This Rule waits for the task to settle, then lets you know when no successful test, lint, typecheck, or build was observed after the latest change. It never runs checks or changes code on its own.
When a coding turn ends without a successful test, lint, typecheck, or build, ClevAgent flags the missing check. You can ask the same agent to run the narrowest relevant verification.
Mishandled secrets
API keys, access tokens, private keys, card numbers, and credential files can leak when an agent reads, writes, or sends them.
ClevAgent warns you when it detects a recognized secret or sensitive credential path. When Security Mode is enabled, it blocks supported API keys, tokens, and private keys before they reach the provider, and pauses high-confidence private credential reads before they run.
Content that tries to hijack the agent
Fetched pages and other supported external tool results can contain instructions that try to override your task or redirect the agent.
When enabled, ClevAgent blocks matching external content before it reaches the model. It tells you what happened and lets you reveal a short, secret-redacted excerpt of the blocked text.
Cleanup that could erase recent work
Supported Git cleanup and reset commands can permanently erase work that has not been saved in version history.
When enabled, ClevAgent runs a read-only Git check and pauses only when it can prove the command would remove uncommitted work. It shows the total count and up to 12 project-relative filenames, never their contents.
The agent changing its own settings
Agent settings files can change what Claude or Codex can access, connect to, or do automatically in future sessions.
When enabled, ClevAgent pauses recognized writes to protected Claude and Codex settings, rule, agent, and MCP files. It shows the affected paths, never their contents.