#1219 Guide agents to inspect reports after failed model and workflow runs

Problem

Reports run after model methods and workflow steps and receive failure-specific context such as executionStatus, errorMessage, workflow status, and failed/skipped step details. However, the generated AGENTS.md guidance only mentions reports for reusable analysis pipelines, while the troubleshooting guide routes failed commands through stderr, --json, tracing, and source inspection without directing agents to inspect generated reports. As a result, agents do not naturally use available report artifacts when diagnosing failed model or workflow runs.

Proposed solution

1. Add a rule to the swamp-managed AGENTS.md section instructing agents to inspect generated reports after a model method or workflow failure, before retrying or changing definitions.
2. Add report inspection to Tier 2 of the troubleshooting skill, after reading stderr and structured command output and before escalating to tracing or source inspection.
3. Include the supported discovery/retrieval commands, preferably directing agents to swamp help report get for current syntax and swamp report get ... --json for structured diagnostics.

Suggested guidance:

Inspect reports after failures. When a model method or workflow run fails, inspect its generated reports before retrying or changing definitions. Reports may contain structured diagnostics derived from execution status, error messages, outputs, and failed steps. Use swamp help report get to confirm retrieval syntax.

Alternatives considered

Users can add repository-local instructions manually, but the relevant AGENTS.md section is swamp-managed and this diagnostic behavior should be consistent across swamp repositories. The report and troubleshooting skill documentation could be updated alone, but agents rely heavily on the generated top-level rules, so updating both locations should make the behavior reliable.