resolve-open-loops

Resolve Open Loops

Use this skill to work through active open_loop and assistant_goal memories in Lore for the current repository, so the stabilisation-guard stands down and the session can proceed cleanly.

Use this skill when

• The stabilisation-guard fired during onSessionStart or blocked a tool call with a message ending in "use /resolve-open-loops to work through these items".
• The user explicitly asks to review, triage, or close open loops for the current repo.
• The user wants a pre-session hygiene pass before starting a large piece of work.

Do not use this skill when

• The goal is to browse or search Lore generally (use memory_search directly).
• The goal is to onboard a new repo's history into Lore (use lore_onboard).
• The session is already unblocked and the user wants to start implementation immediately.

Inputs to gather

Required before starting

• Confirmation that Lore is available in the current session (check memory_status if unsure).
• The current repository — Lore's memory_search scopes results to it automatically.

Helpful if present

• The guard's count from the session-start message (tells you how many items to expect).
• Any indication from the user whether items in a particular area are already done.

First move

Call memory_search for both types in parallel and present the combined list before asking the user anything:

memory_search  query="open loops and pending goals"  type="open_loop"     limit=50
memory_search  query="open loops and pending goals"  type="assistant_goal" limit=50

If both searches return zero results, tell the user the guard should not fire next session and stop — there is nothing to triage.

Routing boundary

Situation	Use this skill?	Route instead
Guard fired — "X active items, use /resolve-open-loops"	Yes	—
User says "let's clear the Lore backlog before we start"	Yes	—
User wants to search what Lore knows about a topic	No	memory_search directly
User wants to recall recent session history	No	lore_recall
Lore is unavailable or not initialised	No	surface the error to the user

Workflow

1. Retrieve active items

Call memory_search twice in parallel — once for each type — to surface the full active list for the current repo:

memory_search  query="open loops and pending goals"  type="open_loop"   limit=50
memory_search  query="open loops and pending goals"  type="assistant_goal"  limit=50

Present the results as a numbered list grouped by type. For each item show: id, content summary (first 120 chars), and scope (repo / global).

2. Triage each item with the user

Work through items one at a time (or in small thematic groups). For each, ask what resolution applies:

Outcome	Action
Done — the work was completed	memory_forget id=<id> supersededBy="resolved: <brief outcome>"
Superseded — a newer item replaces it	memory_forget id=<id> supersededBy="<newer-id or description>"
Still open — genuinely unfinished	Leave it; acknowledge it in context so the user can decide to tackle it now
Invalid — stale or wrong scope	memory_forget id=<id> supersededBy="stale: removed during triage"

Never bulk-forget all items without user confirmation. Resolve one item (or one confirmed batch) at a time.

3. Confirm guard will stand down

After the triage pass, recheck the count:

memory_search  query="open loops and pending goals"  type="open_loop"   limit=50
memory_search  query="open loops and pending goals"  type="assistant_goal"  limit=50

If the remaining count is zero, tell the user the guard will not fire next session.
If items remain, tell the user how many are left and that the guard will still warn — the user decides whether to continue.

4. Optionally save new memories

If the triage revealed something worth retaining (a resolved commitment, a pattern that kept recurring), offer to save it:

memory_save  content="<what was learned>"  type="commitment"  scope="repo"

Only save when the user agrees — do not create memories speculatively.

Guardrails

• Never forget a global-scope item without explicit user confirmation — global memories affect all repositories.
• Never bulk-forget in a single call without the user reviewing the list first.
• Keep supersededBy human-readable — a terse phrase like "resolved: deployed to prod" beats a raw UUID.
• If memory_search returns Lore unavailable (extension not loaded, db locked), surface the error and stop — do not attempt direct DB access.
• Items with scope=transferable should be treated as read-only during this skill; they were written from another repo and the user should resolve them in context.

Validation

Run node skills/skill-authoring/scripts/validate-skill-library.mjs skills/resolve-open-loops/SKILL.md.

Smoke test:

• should trigger: "The guard blocked me again — let's clear those open loops."
• should trigger: "Before we start, let's do a quick Lore hygiene pass."
• should not trigger: "Search Lore for everything we know about the auth module." (→ memory_search directly)
• should not trigger: "Onboard this new repo into Lore." (→ lore_onboard)

Examples

• "The guard fired with 12 items. Walk me through them so we can clear the ones that are done."
• "I want to do a pre-session sweep of my open goals before starting the refactor."
• "Half these open loops are stale — help me triage and mark the dead ones."

Reference files

• references/memory-types.md — memory type definitions, memory_search output format, memory_forget call shapes, and recommended supersededBy phrases for each resolution outcome