SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsae-things-review

ae-things-review

Office & productivity

View repository

Updated 10-16 21:56

About this skill

Item progress review

Follow the evidence-based review method of aha-links, using TE space permissions and the three states plus independent archiving. First read Item operations and collection contract. Do not guess item, space, or owner IDs.

When this round's bound items are involved, first follow execution snapshot and real-time context to read this execution snapshot, then collect the latest review or periodic facts according to the following process; state the two sources separately. When there is no binding, use only the scope explicitly specified by the user.

Read facts

ae-cli things +detail --agent-space-id <space-id> --id <thing-id> --format json
ae-cli things +digest --agent-space-id <space-id> --format json
ae-cli things +digest --agent-space-id <space-id> --thing-ids <id-a>,<id-b> --format json

Scope supports only the current space, owner, or an explicit item set. Do not bring in watchlists or Feishu group scopes. For large scopes, screen first, then deeply review at most 20 items per batch, and disclose the screened count and deeply reviewed count.

Must verify fetchedCount、truncated、consistency、collectedAt. Missing fields are considered contract incompatibility. When truncated or unreadable, draw conclusions only about the facts covered, and do not call the first 100 items the full set.changed_during_collection When , re-read the affected items;non_transactional is still not a strict snapshot of a given moment. Unread attachments, private history, and missing initial goals are explicitly disclosed.

Review order

  1. Goal and acceptance: whether it is clear, reasonable, and verifiable; use the earliest created record to compare with the current goal. If it cannot be reconstructed, review only the current goal.
  2. Scope drift: whether goal changes have a clear record; do not treat later conclusions as the original goal.
  3. Valid updates: results, changes, blockers, or clear next steps are progress. Progress records are candidates pending verification; placeholders, repeated statuses, notes, and title-only changes do not reset the reminder cadence; collected counts are only candidates, and the model still needs to verify whether the body is valid.
  4. Ownership: whether the owner and collaborators are available, and whether the description and progress make the next step clear.
  5. Completion evidence: whether the artifact is accessible and meets acceptance; unread files cannot be claimed to have passed acceptance.

Prefer an explicitly agreed cadence. open means not completed, without distinguishing not yet started from in progress; when there is no agreement, report only the last recorded time and do not invent stagnation. paused checks the pause reason and resumption conditions in the records. Write only "no update recorded," and do not infer "no progress."

Referenced times must include the time zone; when determining natural days according to Asia/Shanghai, first correctly convert the ISO Z/offset, and do not simply remove the suffix. For example, 2026-09-10T04:50:00Z corresponds to Shanghai time 12:50. If conversion is impossible, keep the original UTC value and disclose the limitation.

Output and boundaries

Give healthy / needs attention / high risk / insufficient information, with the goal baseline, evidence (item links and dates), impact, confidence, and at most three priority actions. For workspace reviews, cover common issues first, then summarize item by item; do not evaluate individual performance and do not create an overall score.

By default, only diagnose; do not change status or owner. For specific modifications the user explicitly requests to implement, re-read the version and then execute, and do not repeatedly ask for the same authorization; for inferred changes, first give suggestions pending confirmation. Do not automatically send messages or create scheduled tasks.

File evidence: +detail/+digest automatically paginates to collect asset associations; files with availability=missing or verification_failed are not counted as usable deliverables; even if available, it only means the reference is usable, not that the body has been read or acceptance has passed. Disclose unread files and non-transactional collection boundaries. Private session sources are read only when the current user has permission; the report cites shared matter facts and must not automatically make private history public.