OutlineDriven Fresh reader review
Can a new reader understand the artifact without the context that produced it?
Contract
| Field | Bound contract |
|---|---|
| Trigger | User asks 'does this make sense to someone new' or 'cold-read this', or a README, document, skill, or PR description is about to ship. |
| Authority | Read-only. No file, VCS, credential, paid, published, deployed, or remote mutation. |
| Side effect | One or more isolated clean-room sub-sessions are dispatched depending on artifact stakes and size; the verdict and ordered fixes return in chat and the artifact is never edited. |
| Done | Blind sub-sessions, each reading only the inline artifact with no surrounding context, return a standalone verdict and severity-ordered fixes. |
Inputs
- Artifact (required): The file, document, or text to cold-read. Supplied inline as content, not as a repo path.
- Intent note (required): A one-line private note on what the artifact is meant to be and who it is for. The reviewer never sees this.
Procedure
- Pin scope and private intent. Identify the artifact in focus. Privately note in one line what it is meant to be and who it is for; the reviewer never sees this. Done when: the artifact is identified and the intent is privately noted.
- Segment the artifact if oversized. If the artifact exceeds what one sub-session can hold in context, split it into independent segments at natural boundaries (sections, chapters, files). Each segment gets its own blind read in step 3. Done when: the artifact is either confirmed as a single read or split into segments with boundaries marked.
- Dispatch the required number of blind sub-sessions. Each sub-session receives the artifact (or one segment) inline, not a repo path. Instruct every sub-session: (a) do not open the project's README, docs, or neighbors; (b) read only what is provided inline; (c) diagnose, do not fix. The number of sub-sessions is:
- One for a standard artifact with normal visibility.
- Multiple independent for a high-visibility artifact, irreversible publish, or safety-critical content. Use at least three; each gets the full artifact independently.
- One per segment for an oversized artifact that was split in step 2. Each sub-session reads only its segment. Done when: the required sub-sessions are dispatched with the artifact inline and isolation instructions.
- Collect blind understanding. Ask each sub-session to report what it takes the artifact to be, what is unclear or assumed but unstated, and what it had to guess before it could act. Done when: every sub-session reports its understanding, unclear points, and guesses.
- Compare blind understanding against intent. For each sub-session, compare its blind understanding against the intent noted in step 1. Every mismatch is a defect in the artifact. Done when: every mismatch is identified as a defect.
- Merge, deduplicate, and order defects. When multiple sub-sessions ran, merge their defect lists, deduplicate overlaps, and order by how badly each blocks a fresh reader. A defect reported by multiple independent sub-sessions ranks higher than one reported by a single sub-session. Done when: defects are ordered by severity with concrete fixes, or the verdict 'stands on its own' is returned with an empty fix list.
Failure and recovery
- Context leaked: Abort the affected sub-session. Re-dispatch with explicit instruction to read only the inline artifact and no surrounding files.
- Empty artifact: If the artifact contents are empty or missing, stop and return
blockedwith reasonempty-artifact. No sub-session is spawned. - Sub-session cannot determine artifact type: Report this as a defect (the artifact fails to declare its own purpose). Continue with the cold read.
- No defects found: Return the verdict 'stands on its own' with an empty fix list. Do not invent defects to fill the report.
Output
A report containing:
- Verdict: one of 'stands on its own', 'minor gaps', or 'needs work'.
- Blind summary: what the sub-session(s) understood the artifact to be.
- Defect list: ordered by severity, each with the specific passage and a concrete fix. When multiple sub-sessions ran, note how many independently flagged each defect.