memory-consolidation

Backed by: Node fs primitives + state/lib/log.ts for the eval row.

Cron job (daily 6 AM). Reads all memory files, deduplicates, prunes stale entries, and merges related memories into consolidated entries.

Steps

1. Read all .md files in the project memory directory.
2. For each memory, check if it's stale (references things that no longer exist).
3. Merge memories that cover the same topic into a single entry.
4. Remove duplicates.
5. Write updated files back.
6. append("memory-consolidation", { ok, deduped, pruned, merged }) via state/lib/log.ts.

Eval

Score 1.0 if memories were audited and any stale entries removed. Score 0.5 if audited but no changes needed. Score 0.0 if the memory directory couldn't be read.

Rubric

criteria:
  - name: log_entry_exists
    kind: deterministic
    check: "A log entry for 'memory-consolidation' exists in 'state/lib/log.ts' with fields {ok, deduped, pruned, merged}."
  - name: stale_entries_removed
    kind: judge
    check: "Auditor verifies that stale memory entries (referencing non-existent topics or files) have been removed from the '.md' files in the memory directory."
  - name: duplicates_removed
    kind: judge
    check: "Auditor verifies that duplicate memory entries have been removed, resulting in unique content."
  - name: related_memories_merged
    kind: judge
    check: "Auditor verifies that related memory entries covering the same topic have been merged into a single, consolidated entry."

memory-consolidation - loader

Per-turn rules for the memory-consolidation skill. Full reference: state/skills/memory-consolidation/SKILL.md. Do not skip these.

Critical Rules

• ALWAYS audit memories before claiming pass - score is 1.0 only if stale entries were actually removed
• A clean audit with no changes needed is 0.5, not 1.0 - the score asks "did the loop produce delta" not "did it run"
• Cron-scheduled (daily 6 AM) - don't fire manually unless investigating drift

Commands

| ui dashboard | state/skills/memory-consolidation/resources/ui.openui | |invoke: skill is prose - follow steps 1-6 in state/skills/memory-consolidation/SKILL.md |backed by: Node fs primitives + state/lib/log.ts for the eval row |eval log: state/log/evals.ndjson (skill: "memory-consolidation") |append helper: append("memory-consolidation", { ok, deduped, pruned, merged })

OpenUI Resource

• Skill-owned OpenUI Lang resource: state/skills/memory-consolidation/resources/ui.openui. Read it before rendering or editing this skill's generated component surface.
• Treat this resource as a first-class artifact of the skill, not a generic chat response. Improve it when the skill's user-facing output needs to become richer.
• System resources compose OpenUI primitives and inherit SnappyChat tokens. Use ui_contract: branded in SKILL.md only for deliberate platform or client visuals.

Known Pitfalls

• "Stale" means references to things that no longer exist - verify before pruning, deletion is destructive
• Score 0.0 only if the memory directory could not be read; a successful read with no changes is 0.5

Self-Test

An agent reading this should correctly:

1. [ ] Score 0.5 (not 1.0) when audit completes with zero changes needed?
2. [ ] Verify a "stale" reference no longer exists before pruning?
3. [ ] Append the run via append("memory-consolidation", ...) from state/lib/log.ts?

Self-report

If this loader fell short, append a line:

echo "[$(date -u +%FT%TZ)] memory-consolidation: <what was missing>" >> state/log/loader-feedback.log

<!-- kernel-ok: historical attribution for the PID-loader pattern. -->

Self-correcting loader (PID feedback)

This is the rule that makes loaders learn. Ported from snappy-kernel, where it demonstrably worked: the snappy-dom-cartographer skill mapped Skool admin UI once, every later ship-agent inherited the full map for free.

Before you finish, do two things:

1. Fix gaps (P - proportional). MANDATORY. If this loader didn't cover your case - if you had to read another file, run an undocumented command, work around a wrong selector, or discover a quirk - you MUST attempt an Edit to this AGENTS.md before you log.

• Default action: edit this file. Surgical. Add the missing line, table

row, or code example where it belongs. No restructuring.

• Only LOGGED is allowed when: the fix needs >10 lines, spans multiple

files, or requires a structural rewrite. In that case the state/regen/drain.sh queue picks it up asynchronously.

• "I didn't have time" / "it's minor" / "the next agent will figure it out"

are NOT valid reasons. Minor gaps compound - that's the whole point of this loop.

• Concrete example of a valid P-fix: discover agent-browser screenshot

takes path positionally not as --output <path>. One line in a table. Edit it in. Do not just log it.

• The goal: the next agent never has to leave the loader.

2. Log the result.

echo "[$(date -u +%FT%TZ)] memory-consolidation: <what was missing or fixed> [FIXED|LOGGED]" >> state/log/loader-feedback.log

• FIXED = you patched this loader inline (P-fix).
• LOGGED = too large for inline; the PostToolUse enqueue + Stop-hook drain

will rewrite the loader from scratch on next session-end.

Do not skip this. Every agent run must leave the system better than it found it. The loader is the setpoint; you are the sensor; the gap is the error signal; closing the gap is the correction.