confidence-fragility-review
Review whether an artifact's confident claims are backed by evidence, tests, contracts, or operational safeguards. Use before shipping, publishing docs, accepting generated plans, or trusting a workflow that may present fragile assumptions as certainty.
Install
npx skills add https://github.com/markoblogo/abvx-agent-skills/tree/main/skills/confidence-fragility-review
claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install markoblogo-abvx-agent-skills@llmmart
git clone https://github.com/markoblogo/abvx-agent-skills.git
The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole markoblogo/abvx-agent-skills collection as a plugin from our marketplace. Git is the plain clone.
Skill manifest
Confidence Fragility Review
Stress-test the gap between how confident an artifact sounds and how much evidence actually supports it.
Use When
- a plan, release note, README, landing page, or SET bundle sounds ready but proof is thin;
- generated docs imply stronger guarantees than the repo can support;
- final review needs a fear-oriented lens before ship;
- a user asks what could go wrong if people trust the artifact at face value.
Workflow
- Identify confident claims:
- readiness claims;
- safety claims;
- completeness claims;
- compatibility claims;
- performance or reliability claims.
- For each claim, ask what backs it:
- tests or CI;
- schema or type contracts;
- runtime checks;
- docs links;
- human approval;
- explicit limitations.
- Classify the concern:
- tactical: a specific path can fail;
- structural: a category of failure has no defense;
- epistemic: confidence is not earned by evidence.
- Separate justified fragility from projected worry.
- Recommend the smallest evidence step that would make confidence warranted.
Output Shape
Use:
- verdict:
CONFIDENCE_WARRANTEDorFRAGILITY_MASKED; - confidence claims reviewed;
- evidence found;
- fragility register: tactical, structural, or epistemic;
- smallest proof step.
Guardrails
- Do not turn this into a generic risk audit.
- Do not demand proof for every harmless claim.
- Do not perform security analysis unless the artifact makes security claims.
- Be explicit when a fear is plausible but not evidenced.
Provenance
Adapted from the anxiety-reader pattern in aself101/agents-and-pipelines, reframed as a compact confidence-vs-evidence review skill.
Files (abvx-agent-skills)
-
agents
-
openai.yaml 212 B
Loading…
-
-
SKILL.md 2 KB
--- name: confidence-fragility-review description: Review whether an artifact's confident claims are backed by evidence, tests, contracts, or operational safeguards. Use before shipping, publishing docs, accepting generated plans, or trusting a workflow that may present fragile assumptions as certainty. license: MIT metadata: abvx_status: experimental abvx_origin: adapted --- # Confidence Fragility Review Stress-test the gap between how confident an artifact sounds and how much evidence actually supports it. ## Use When - a plan, release note, README, landing page, or SET bundle sounds ready but proof is thin; - generated docs imply stronger guarantees than the repo can support; - final review needs a fear-oriented lens before ship; - a user asks what could go wrong if people trust the artifact at face value. ## Workflow 1. Identify confident claims: - readiness claims; - safety claims; - completeness claims; - compatibility claims; - performance or reliability claims. 2. For each claim, ask what backs it: - tests or CI; - schema or type contracts; - runtime checks; - docs links; - human approval; - explicit limitations. 3. Classify the concern: - tactical: a specific path can fail; - structural: a category of failure has no defense; - epistemic: confidence is not earned by evidence. 4. Separate justified fragility from projected worry. 5. Recommend the smallest evidence step that would make confidence warranted. ## Output Shape Use: - verdict: `CONFIDENCE_WARRANTED` or `FRAGILITY_MASKED`; - confidence claims reviewed; - evidence found; - fragility register: tactical, structural, or epistemic; - smallest proof step. ## Guardrails - Do not turn this into a generic risk audit. - Do not demand proof for every harmless claim. - Do not perform security analysis unless the artifact makes security claims. - Be explicit when a fear is plausible but not evidenced. ## Provenance Adapted from the `anxiety-reader` pattern in `aself101/agents-and-pipelines`, reframed as a compact confidence-vs-evidence review skill. -
SKILL_CARD.md 1.4 KB
Loading…
Comments (0)
Sign in to join the conversation.
Reviews (0)
No reviews yet.
No comments yet.