answer-all-open-questions-with-recommendation

answer-all-open-questions-with-recommendation

This is the batch path that records every open question's embedded recommendation as its answer. The recommend sweep (/recommend-all-open-questions) annotates each question by embedding a <recommendation> element in its <open-question> block; this skill re-checks the current milestone's questions and, for each block that contains such an element, dispatches the file-editing answer-open-question-with-recommendation agent to lift that recommendation and record it — leaving recommendation-less blocks untouched (annotating them is the recommend sweep's job, not this one's). It requires no clean-working-tree precondition.

It is an orchestrator. It does not record answers itself: for each recommendation-bearing question it dispatches the agent, which lifts the <recommendation> element, folds the decision into ## Decisions, cascades to mooted siblings, and leaves that edit staged but uncommitted. The orchestrator commits that staged index itself — once per successful agent return, before dispatching the next. Because every dispatch mutates the same requirements.md, the dispatches run strictly sequentially, never in parallel.

Usage

/answer-all-open-questions-with-recommendation

Takes no arguments — it sweeps every <open-question> block (both status="open" and status="deferred") in the current milestone's requirements.md that contains a <recommendation> element.

Workflow

0. Find the current milestone

Follow .agents/plugins/cairn/shared/get-current-milestone.md to resolve <MILESTONE_DIR>. Never use a hardcoded task-list path.

1. Gather the recommendation-bearing questions once and order them by the dependency graph

Fetch the recommendation-bearing set with the line-oriented boundary-line CLI — do not read the whole file to eyeball headers. Every <open-question> block lives under the single ## Open questions section of <MILESTONE_DIR>/requirements.md, so that section is the one bounded region the CLI slices deterministically. Using awk/sed/grep keyed on the <open-question …> opening and </open-question> closing boundary lines — never a real XML processor (xmllint) — enumerate every block in document order (the order the CLI hands them back is the order their blocks appear in the section, and the walk below relies on it) and keep only those that contain a <recommendation> element, extracting each surviving block's id (and status) by attribute-name-anchored regex like id="([^"]*)" (so extraction is independent of attribute order). From each surviving block also extract the question value of every self-closing <depends-on question="…" option="…"/> line it carries — the same attribute-name-anchored regex, anchored on question=; the recommend sweep embeds one such line per sibling recommendation this block's recommendation assumed. Read only the question value here: the option value is what the recording core's cascade reconciles at answer time, and the orchestrator never acts on it. Open and deferred blocks share one <open-question …> / </open-question> boundary-token pair distinguished only by status, so the gather ignores type. Recommendation-less blocks — those with no <recommendation> element — are not gathered: annotating them is the recommend sweep's job (/recommend-all-open-questions), never this one's. If no block carries a <recommendation> element, say so and stop.