ai-agent-case-study
Skill1 starsChanged 4 months agoMITView on GitHub
Structure and write design case studies for AI agents, copilots, autonomous workflows, multi-agent systems, and other probabilistic products. Use when a case study must explain autonomy levels, permissions, approval thresholds, steering, observability, user-visible rationale, uncertainty, trust calibration, streaming, latency, orchestration, failures, retries, rollback, state recovery, evals, or human-in-the-loop design.
In this repository11 other files it gives its agents.
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 4 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 8
Treat the agent as a probabilistic collaborator operating inside a control system.
- Distinctive: line 14
Never request or publish hidden chain-of-thought.
- Very specific: line 18
Read references/supervise-framework.md, then cover:
- Very specific: line 30
Use assets/ai-agent-case-study-template.md for a new draft.
- Distinctive: line 45
Avoid generic process theater.
- Distinctive: line 52
when it acts silently versus requests approval.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
AI Agent Case Study
Treat the agent as a probabilistic collaborator operating inside a control system. Do not present it as a linear CRUD flow or a magical chat box.
Start with evidence
Request or inspect the architecture, tool and permission model, interaction states, approval rules, traces, action history, evals, incidents, metrics, research, screenshots, and decision records. If an artifact is missing, mark the gap.
Never request or publish hidden chain-of-thought. Use user-visible rationale, action summaries, provenance, tool calls, diffs, logs, confidence signals, and outcomes.
Apply SUPERVISE
Read references/supervise-framework.md, then cover:
- System role and scope
- Uncertainty model
- Permissions and autonomy
- Execution observability
- Recovery and reversibility
- Verification evidence
- Intervention mechanics
- Success telemetry
- Evolution and learning
Use assets/ai-agent-case-study-template.md for a new draft.
Required distinctions
- Separate model behavior, orchestration behavior, interface behavior, and product outcome.
- Separate designed safeguards from safeguards proven in execution.
- Separate a confidence display from actual calibration evidence.
- Separate system completion from user success.
- Separate latency perception from raw model latency.
- Show both the autonomous path and the human intervention path.
Narrative order
Lead with the supervisory problem and pivotal control decision. Then explain the system boundary, human-agent interaction model, observability, failure recovery, trade-offs, validation, outcomes, and limitations.
Avoid generic process theater. Include interviews or personas only when they changed a control, trust, or system decision.
Done
The reader can reconstruct:
- what the agent may perceive, decide, and change;
- when it acts silently versus requests approval;
- how the user pauses, steers, corrects, rejects, or reverses work;
- what the interface reveals during long-running or failed execution;
- which signals calibrate trust without implying certainty;
- what evidence proves the claimed outcome.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

