trajectory-review
SolidPost-hoc diagnosis of a failed agent trajectory. Classifies the first unrecoverable step into one of nine failure categories (plan adherence, hallucinated information, invalid tool call, misread tool output, intent–plan mismatch, under-specified intent, unsupported intent, guardrail trigger, system failure) and produces an evidence-backed root-cause report.
Install
Quality Score: 82/100
Skill Content
Details
- Author
- majiayu000
- Repository
- majiayu000/vibeguard
- Created
- 6 months ago
- Last Updated
- today
- Language
- Shell
- License
- MIT
Similar Skills
Semantically similar based on skill content — not just same category
agent-trajectory-evaluator
Evaluate a multi-step AI agent's whole run — tool calls, intermediate steps, and final result — not just final-answer correctness, so you can pinpoint WHERE it went wrong. Use when building or debugging a tool-using or multi-step agent, when final-answer-only evals can't explain failures, or when a prompt/model change quietly makes the agent less efficient or more error-prone even though the answer still looks right.
trajectory
Discover, run, interpret, and iteratively tune a project's existing agent trajectory evaluation capability. Use when the user asks to evaluate agent trajectories, compare effect and cost across runs, find trajectory smells or the next trajectory problem, or identify optimizations for a known trajectory issue. Operate project-owned trajectory evidence and canonical entrypoints; do not invent missing evidence, verdicts, or a parallel evaluation framework.
agent-failure-diagnosis
Diagnose why an AI agent behaved badly, using operationalised criteria that two independent reviewers can apply to the same evidence and reach the same answer. Use this whenever someone describes an agent that misbehaved, went off track, did something unexpected, ignored instructions, made things up, went beyond its scope, got stuck in a loop, lied about what it did, or "went rogue" — and whenever reviewing agent traces, logs, or transcripts to work out what went wrong. Use it for post-incident analysis, for design reviews asking "how could this fail?", and when someone needs to classify agent failures consistently enough to spot patterns across many incidents. Reach for this even when the user just wants an explanation rather than a formal report, because the classification is what makes the explanation defensible later.