discard

Discard

Remove a proposed staging directory. The live contextualizer is not modified. Idempotent — discarding a non-existent proposed dir exits cleanly with a one-line diagnostic.

When to invoke

Invoke /skill-engine:discard <name> when reviewing a proposal that does not warrant promotion — the predictions in Step 1 of REVIEW.md and the disagreement set in Step 2 surface enough divergence that re-running DISCOVER with a --hint=... is the right next move, not promotion. discard is also the path out when the user ticked [x] reject in Step 3: /skill-engine:apply refuses to promote a rejected proposal and points the user here.

Resolving <name> and <install>

Same resolution as /skill-engine:review. See review/SKILL.md § Resolving <name>.

Pre-discard prompt

Before removing anything, render a one-line confirmation naming the path:

Remove <install>/<name>-context.proposed/ (and its .review/ audit trail)? The live <name>-context/ is untouched. [y/N]

Accept y or yes (case-insensitive, leading/trailing whitespace trimmed) as consent. Treat N, blank input, or anything else as decline — the prompt is a single-shot, no re-prompt.

On decline, exit cleanly with:

Discard cancelled. <install>/<name>-context.proposed/ is untouched.

The removal

On consent:

rm -rf "<install>/<name>-context.proposed/"

Exit message:

Removed <install>/<name>-context.proposed/. Live <name>-context/ is untouched.

Idempotent paths

When the proposed dir does not exist at the resolved path (e.g., a stale invocation after an earlier discard, or a slug typo that resolves nowhere), surface:

No proposed staging dir at <install>/<name>-context.proposed/. Nothing to discard.

Exit zero (not an error condition — discarding nothing is a no-op).

When bare invocation finds zero *-context.proposed/ directories under <install>, same shape:

No proposed staging dirs found under <install>. Nothing to discard.

Exit zero.

What this skill does NOT do

• It does not touch the live contextualizer.
• It does not preserve a backup or move the proposed dir to a trash location — rm -rf is the operation. If the user wants to keep the REVIEW.md audit trail from a discarded proposal, they should copy it out themselves before invoking discard.
• It does not unstage upstream-source clones from ~/.cache/skill-engine/. The cache is independent of the proposed dir; /skill-engine:clean-cache is the cache-management surface.
• It does not require sign-off in REVIEW.md. Discarding a proposal is the cheapest path out and should not be gated on the user having gone through the predict-then-compare loop.