mainframe-harness-feedback

Report a MAINFRAME harness fault

Use this skill when available evidence indicates that MAINFRAME, an adapted MAINFRAME component, or the effective agent harness behaved incorrectly. Keep this feedback separate from the receiving project's issue queue.

First distinguish the owner of the problem:

• Use the receiving project's issue route for an application, data, infrastructure, or project-documentation defect.
• Repair a local project-harness configuration defect when that repair belongs to the assigned work and is authorized.
• Use this route when the evidence concerns MAINFRAME semantics, adaptation, discovery, precedence, registration, hooks, skills, agents, commands, permissions, or behavior across supported invocation trajectories.

Do not require a proven root cause before preserving a useful observation. State separately what was observed, what was reproduced, and what remains an inference.

Use the resolved MAINFRAME destination

Use only the configured feedback destination:

{{MAINFRAME_ROOT}}/docs/tickets/open/observations/

Treat an unresolved placeholder as an installation defect. Do not scan the home directory, other repositories, mounted volumes, application data, archives, backups, recovery directories, or Trash to guess the root. If the resolved root or the narrow feedback permission is unavailable, return feedback-ready content and the exact missing path, capability, or authorization to your immediate caller; do not write the report into the receiving project's queue or claim it was filed.

Use only the access needed to list open observations and create or update a matching record. Do not use the feedback permission to inspect or modify unrelated MAINFRAME content.

The destination may be absent before the first report. Create only the exact destination directory when writing the first real observation and the configured authority permits it.

Preserve useful evidence

Record:

• a concise factual title;
• the affected canonical or adapted component;
• the receiving agent product, version, and interface when known;
• the execution trajectory relevant to the fault, such as primary agent, subagent, reviewer, direct request, delegated task, Desktop, or CLI;
• the triggering action or condition;
• observed behavior and expected behavior when a source defines it;
• the practical consequence for the active work;
• minimal redacted evidence and reproduction status;
• remaining uncertainty.