- Home
- Skills
- Agents & Orchestration
- Agentic Run Failure Forensics
Works with the AI Agents you already use
Agentic Run Failure Forensics
Trace multi-agent failures, separate roots from symptoms, and design the smallest evidence-producing rerun.
$19
Agentic Run Failure Forensics
Example session with this skill installed
Objective: Add CSV export and prove it passes tests. Plan v2 requires an independent reviewer to inspect the diff and run pytest tests/test_export.py. Implementer: “CSV export is complete. I did not run tests.” Handoff: “All requirements satisfied.” Reviewer: “Approved based on the implementer report.” Ledger: tests=complete. No test result supplied; export.py is untracked. Why did this multi-agent run fail?
- Read your context and instructions
- Compiled the agentic run failure
EVIDENCE COVERAGE — Plan and roles: PRESENT. Test execution: MISSING. Repository state: PRESENT and conflicts with completion. SUPPORTED FINDING F-01 — FM-3.2 No or Incomplete Verification. Evidence: implementer says tests were not run; reviewer approved from self-report; no test record exists. COMPLETION CONTRADICTION — ledger tests=complete is an assertion, not proof. ROOT/CASCADE CANDIDATE — missing executable gate → unsupported approval → false completion record. MINIMAL RERUN — freeze the diff, run the named test in a clean state, capture result and git status, and block completion when evidence is absent.
Connects securely to your tools. The creator never sees your data.
About this skill
Your agents finished. The evidence says the run did not.
Agentic Run Failure Forensics turns plans, agent traces, handoffs, tool effects, progress ledgers, review reports, tests, git state, and final artifacts into an evidence-bound failure report. It distinguishes observed facts from agent claims, maps supported behavior to the published MAST taxonomy, exposes stale ledgers and false-completion signals, and proposes the smallest rerun that can test a repair hypothesis.
What it produces
- Evidence coverage across objective, plan, roles, handoffs, tool effects, verification, repository state, and final result.
- Intended-versus-observed run reconstruction without filling missing steps.
- Supported MAST failure classifications with exact evidence pointers and alternatives.
- Root/cascade map separating entry conditions, contributors, propagation, symptoms, and impact.
- Ledger and completion contradictions, including reviewer passes that conflict with tests or artifacts.
- Repair layer selection, minimal rerun plan, and a regression fixture.
Why it is different
This is not a generic log summary or another debugging prompt. Dedicated reference protocols govern MAST classification, artifact normalization, evidence status, causal restraint, cascade mapping, repair selection, privacy, and injection isolation. Twelve forward cases cover complete and partial traces, single-agent exclusion, role conflict, handoff loss, premature termination, false reviewer passes, stale ledgers, embedded injection, multiple entry conditions, tool outages, and secret redaction.
Best for
- Teams using multi-agent coding or research workflows.
- Superpowers-style planning, implementation, review, and verification runs.
- Claude Code, Codex, Cursor, and other agent-host users with inspectable artifacts.
- Engineers investigating loops, lost handoffs, stale state, or unsupported completion claims.
Boundaries
The skill does not debug single-agent prompts, repair code, execute agents, prove causality, certify security, or guarantee that a proposed repair will work. It does not treat MAST as exhaustive. Partial traces produce bounded findings and missing-evidence requests, not reconstructed history.
Independent product: Designed for Superpowers-style and other multi-agent workflows. Not affiliated with or endorsed by obra, Anthropic, OpenAI, Cursor, or any agent framework.
Requires artifacts from at least two agent roles. Partial traces produce bounded findings and missing-evidence requests, not reconstructed history. A causal root is reported only when a transfer mechanism is supported; otherwise the output lists root candidates. It does not repair code or execute reruns. Runtime compatibility was not forward-tested.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 3 months ago
Permissions
No special permissions declared or detected