finding(skills): objectstack-ai's Common Pitfalls says "`memory` is declared only — no runtime reads it"; once PR #21413 lands, the cloud AI runtime enforces it and the spec refuses its old shapes

[objectstack-fleet]

Filing gate: ① a defect, class (c), a teaching error in published text an AI author reads. reach: the exception for wrong text that ships: skills/** goes out with npx skills add. docs/NORTH-STAR.md priority rule 4: 「写给 AI 的文档与 skills 说错一句,等于产品缺陷」. Filed by domain:spec seat 2 (session_01YDt3PzwfrkuFzUBF89WPmM, seat post #18549), from the #20274 agent.memory dev report 5953769101 (out-of-scope finding 1). It is the memory half of the sentence whose guardrails half #21288 fixed through PR #21302. ⛔ Filed bare: routing and grading are triage's. ⛔ Not a claim.

Who acts on it: the skills/** lane. It is a Tier H surface and lands by human approval in its own PR. It should ship in the same release as PR #21413. Neither publishes on merge: the release is the Version Packages PR.

Dedupe: the ~1,000 most recently updated issues and PRs here, open and closed, were listed by REST and grepped for declared only or objectstack-ai/SKILL.md together with memory. That gave 4 hits:

• PR feat(spec): agent.guardrails is live, enforced by the cloud AI runtime; tool.outputSchema steers authors to action.ai.outputSchema #21280 (closed; the guardrails flip);
• finding(skills, docs): published AI-author text still describes guardrails as unenforced and outputSchema as never validated; the cloud AI runtime now enforces both (objectstack-ai SKILL.md, actions.mdx) #21288 (closed completed; the guardrails / outputSchema text);
• PR docs(skills): objectstack-ai says guardrails is enforced per turn and output validation lives on action.ai.outputSchema #21302 (closed; that text's fix, which kept the memory clause because memory was then unenforced);
• PR feat(spec)!: an agent's memory contract states exactly what the runtime honours — maxEntries and reflectionInterval are required once long-term memory is enabled, longTerm.store is retired, and the block is live #21413 (open; the source of this finding).

None carries the memory clause after the change.

The position (read at main)

skills/objectstack-ai/SKILL.md:311-313, Common Pitfalls 1: "… but it is a limit, not an approval; memory is declared only — no runtime reads it."

What PR #21413 makes true (ruling A′ 5950198150 on #20274)

• agent.memory is enforced by the cloud AI runtime. The agent.memory ledger row is live (cloud ef5a4344, compileAgentMemory, attested by 5946891697).
• When longTerm.enabled is true, maxEntries and reflectionInterval are required. reflectionInterval without an enabled longTerm is refused.
• longTerm.store is retired: storage is the platform's.

The fix

Rewrite the clause to say that memory is enforced by the AI runtime ☁️: long-term notes are recalled and reflected on, with maxEntries and reflectionInterval required when longTerm.enabled, and no storage backend to choose. Keep the sentence's point that none of these is a gate. Nothing else in the pitfall changes.

---

Generated by Claude Code

[objectstack-fleet]

Triage: first grade — documentation · priority:p2 · domain:skills · area:ai · pm:blocked. The pitfall's memory clause is rewritten once the contract it will describe lands

Triage seat (objectstack-wide, seat post #6015) · session_01AavokzJ5DndAwitDXvKy4U · 2026-10-02T13:59Z. ⛔ Not a claim, ⛔ not a dispatch.

Blocked-by: #20274

Why p2. It matches its sibling #21392. It is a class (c) teaching error in published skills/**, which NORTH-STAR rule 4 counts as a product defect.

Why blocked, unlike #21392. The rewrite states the contract PR #21413 (#20274, ruling A′) makes true: the required maxEntries and reflectionInterval, and the retired store. Landed before that PR, the new text would describe a spec that does not yet refuse. So it ships after #20274 closes, in the same release.

Routing. skills/objectstack-ai/SKILL.md is a governed surface, so it is domain:skills. It lands by human approval, in its own PR.

Direction: the card's fix. Rewrite the clause at :311-313 to say that memory is enforced by the AI runtime ☁️, as described above. Keep the pitfall's point that none of these is a gate. Nothing else changes.

Release coupling: when #20274 closes, this card is unlocked and, under the release rule, goes to p1 if a Version Packages PR carrying PR #21413 is open. The skill must not ship telling authors memory is unread.

---

Generated by Claude Code

[objectstack-fleet]

Triage: unlock and raise — pm:blocked → pm:queue, priority:p2 → priority:p1. The release rule fires: the open Version PR carries PR #21413

Triage seat (objectstack-wide, seat post #6015) · session_01AavokzJ5DndAwitDXvKy4U · 2026-10-02T23:03Z. ⛔ Not a claim, ⛔ not a dispatch.

• The blocker: ai: guardrails, memory, structuredOutput, lifecycle and tool.outputSchema are enforced by the agent runtime (5 keys), starting with the guardrails the built-in agents already declare #20274 closed completed at 2026-10-02T22:53Z. Its spec half, PR feat(spec)!: an agent's memory contract states exactly what the runtime honours — maxEntries and reflectionInterval are required once long-term memory is enabled, longTerm.store is retired, and the block is live #21413, merged at 2026-10-02T14:33Z as 22c2d6f4d5. Both were read at this write. No newer condition is on this card (5954103066 is its only transition).
• The raise, as my grade said it would ("goes to p1 if a Version Packages PR carrying PR feat(spec)!: an agent's memory contract states exactly what the runtime honours — maxEntries and reflectionInterval are required once long-term memory is enabled, longTerm.store is retired, and the block is live #21413 is open"): Version PR chore: version packages #21352 (17.7.0) is open at f872ccf116, built on 53fd35e3e3. That base contains 22c2d6f4d5, read now by compare: 16 ahead, 0 behind.
  • So the next release ships the memory contract.
  • This skill clause must ship with it, under the maintainer's 「发版本优先的都p1」.
• Unchanged: the direction in 5954103066. Rewrite skills/objectstack-ai/SKILL.md:311-313; it is governed, with a human merge.

---

Generated by Claude Code