Dynamic Workflows

Dynamic Workflows

Dynamic workflow views group recurring agent execution patterns and dashboard-launched runs. They help compare orchestration behavior across sessions without replacing transcript-level evidence.

Questions workflows can answer

• Which sessions use subagents and how deeply?
• Where do retries or compactions recur?
• Which patterns usually complete successfully?
• How many tools, tokens, and agents did a run consume?
• Did a dashboard-launched process finish, fail, or remain live?

Explore from aggregate to instance

npx ccam workflows
npx ccam workflows --session <session-id>
npx ccam runs --session <session-id>
npx ccam run history
npx ccam run get <run-id>

Use aggregate patterns to find a candidate, then inspect the actual session, agent tree, events, and transcript. A detected pattern is an interpretation of recorded behavior, not a guarantee of agent intent.

Comparing workflows fairly

Align project type, provider/model, time range, and task complexity. Token totals alone cannot tell whether one approach is better. Consider correctness, elapsed time, operator intervention, retries, and produced artifacts.

Operational use

Workflow trends can reveal unusually deep fan-out, repeated failed tool sequences, or compaction-heavy sessions. Turn a recurring observation into an alert or team practice only after sampling individual cases.

Run history

Dashboard-launched runs have live handles and persisted history. After restart, treat persisted records as evidence of the previous run, not proof that an OS process remains active. Use supported run controls and verify the project working tree before retrying or resuming.

Related: Running Agents from CCAM and CLI Automation Patterns.