agent-readiness

Agent Readiness

Rates a page for how safely an autonomous agent could act on it directly, with no human in the loop to catch a misreading. It flags ambiguous pronouns, steps that depend on context the page never states, prose that buries instructions instead of structuring them, underspecified actions with real blast radius, and genuinely non-deterministic steps, then outputs a score and a blast-radius-ordered fix list.

Inputs

• {{PAGE}}: The documentation page to rate for agent readiness.

How to apply

Substitute each {{PLACEHOLDER}} above with the user's actual content, then follow the prompt below exactly — including its output format and its stated limits.

---

You are rating how safely an autonomous agent (not a human) could read this page and act on it directly — running commands, editing files, calling APIs — without a human in the loop to catch a misreading.

Checks

Go through PAGE looking for each of these failure modes. For each instance found, quote the exact text.

1. Ambiguous pronouns/references — "it," "this," "that config" where more than one plausible referent exists nearby. An agent will pick one; you're checking whether it could pick wrong.
2. Context-dependent steps — a step that only makes sense if the reader already has state the page never established (a variable value, a working directory, a prior step's output) and doesn't say so.
3. Missing structure — instructions buried in prose paragraphs instead of numbered steps or code blocks, where an agent parsing for actions would have to infer step boundaries.
4. Dangerous underspecified claims — an instruction phrased as an executable action but missing a detail that changes its blast radius ("delete the old key" without which key, "run the migration" without which environment) — the kind of gap a careful human fills with judgment that an agent won't have.
5. Non-determinism — a step with genuinely more than one valid reading, where different correct-sounding interpretations lead to different outcomes.