Long-running PentesterFlow sessions can exceed context limits, forcing compaction. When this happens, execution state is not reliably preserved.

[PNP-MA]

Feature Request: Persistent Task Checkpoints Across Context Compaction

Problem

Long-running PentesterFlow sessions can exceed context limits, forcing compaction. When this happens, execution state is not reliably preserved.

Requested Feature

Add automatic task checkpoints that survive:

• Context compaction
• Agent restarts
• Session continuation
• Long autonomous runs

Suggested Implementation

Store structured checkpoints containing:

• Objective
• Plan
• Current progress
• Completed tasks
• Pending tasks
• Findings
• Evidence
• File references
• Tool outputs

The agent should automatically reload the latest checkpoint after compaction and continue execution without requiring manual intervention.

Benefits

• Reduced duplicate work
• Better autonomous operation
• More reliable long-running assessments
• Improved user experience
• Lower token consumption from repeated analysis

[PentesterFlow]

Update: this is now covered in 181cceb.\n\nThe agent now persists structured session memory across compaction and resume, including objective, plan, completed work, findings/evidence, files, commands, credentials/placeholders, and TODOs. Manual and auto-compaction merge into that memory, resume reloads it into context, and periodic/manual redacted context snapshots are available.\n\nValidated with npm run ci before push.