Learn This Project

This skill helps an agent learn an unfamiliar repository quickly and report only what is supported by evidence in the repo.

The goal is not to read everything. The goal is to produce a reliable working model of the project, identify the key files that support that model, and stay explicit about uncertainty.

When to Use

Use this skill when:

• a task starts in an unfamiliar repository
• the user asks for a repo overview, architecture summary, or onboarding help
• you need to identify conventions before making changes
• you need a quick triage pass before deeper implementation or review work

Do not use this skill as a substitute for task-specific investigation once the relevant subsystem is already known.

Core Objectives

1. Identify the repository's likely purpose, audience, and main capabilities.
2. Map the major components, boundaries, and execution flow.
3. Capture the tech stack, workflows, and engineering conventions that matter

for follow-on work.

1. Separate direct evidence from inference and avoid false certainty.
2. Call out ambiguity or insufficient evidence instead of guessing.

Working Rules

• Prefer primary artifacts such as code, config, tests, manifests, workflows,

and runtime entrypoints over prose docs alone.

• Treat a fast repo triage as intentionally partial unless you actually

inspected the repo broadly.

• Cite concrete files whenever they materially support a claim.
• If the repository is large, sample deliberately instead of wandering.
• If the evidence is weak, say so plainly.

How to Learn the Repo

Start broad, then narrow:

1. Inspect the top-level repository shape.
2. Read project framing such as README*, manifests, Makefile, and CI.
3. Confirm the implementation surface through entrypoints, package boundaries,

service wiring, or exported APIs.

1. Inspect configuration, integrations, and quality signals.
2. Summarize conventions and remaining uncertainties.

For the detailed workflow, sampling strategy, and ambiguity rules, load:

• references/WORKFLOW.md for the step-by-step repo-triage process
• references/EVIDENCE.md for confidence levels, evidence standards, and

insufficient-evidence handling

Output

Produce a concise report that covers:

• project summary
• tech stack and tooling
• architecture and execution model
• conventions and quality signals
• ambiguities and gaps
• confidence notes

Use the output template in assets/report-template.md when you need a concrete response shape. Load it before writing the final report if the user wants a structured project summary or if the repo is complex enough that a fixed format improves clarity.

For more output guidance, load:

• references/OUTPUT.md for section expectations and writing rules
• assets/report-template.md for a fill-in template

Loading Guidance

Do not load every supporting file by default.

• Load references/WORKFLOW.md when you are actively exploring a repository.
• Load references/EVIDENCE.md when the repo is ambiguous, large, stale, or

lightly documented.

• Load references/OUTPUT.md and assets/report-template.md when preparing

the final summary for the user.

Success Criteria

The skill is successful when the final report:

• gives a useful working model of the repository
• distinguishes evidence from inference
• references the files that matter most
• avoids invented details
• tells the reader what is still unclear