agentsclimarketplace

Dreaming

Skill davidlee/doctrine/plugins/doctrine/skills/dreaming

bathe your agents in engineering rigour and flames

Install
npx -y skills add davidlee/doctrine --skill dreaming

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Unified memory corpus maintenance — covers both reactive (change-triggered) and proactive (periodic/idle-time improvement) paths. One skill, two entries.

SKILL.md

3.2 KB, as published. Nobody here has run it

Dreaming

Memory decays. Files move, commands change, invariants shift, duplicates accumulate. This skill is the maintenance loop that keeps the corpus trustworthy — one procedure, entered from either direction.

Reactive entry. Something changed: a file was renamed, a command flag was removed, an ADR shifted a subsystem boundary, a code review found stale memories referencing old paths. Run the procedure scoped to the affected area.

Proactive entry. Idle time, end-of-slice wrap-up, or a scheduled maintenance window. Run the procedure corpus-wide.

Procedure

  1. Validate. Run doctrine memory validate corpus-wide (reactive: scope with --path-scope/--glob to the changed area). Triage each finding:

    • Fix immediately if trivial (update a path, refresh a scope)
    • Capture as a backlog item if non-trivial (risk, chore, or improvement)
    • Note and defer if low-impact and not time-sensitive

  2. Prune. Identify stale memories:

    • Past review_by date → flag for review or archive
    • Unverified threads past expiry → archive (memory status <REF> archived)
    • working-lifespan memories older than a few days → retract or promote to a durable type
    • Run doctrine memory status <REF> archived for stale items; use retracted for memories that were simply wrong.

  3. Link. Strengthen the graph:

    • Review suggested relations on recently-recorded or recently-edited memories (check memory show <REF> for relation rows)
    • Check for orphans — memories with no inbound and no outbound relations memory list --orphans
    • If there are orphans, focus on these - linking 20-30 records this run
    • If there aren't, explore the corpus and strengthen / prune connections between existing memories
      • use memory find <keyword> to find candidate memories for linking
    • Run doctrine link for high-confidence matches

  4. Backlog grooming. Findings that can't be fixed in this pass become backlog items:

    • Risks for not-yet-surfaced issues (e.g., a memory pointing at a deprecated API)
    • Chores for cleanup work (e.g., merging duplicate memories)
    • Improvements for enhancements (e.g., promoting a pattern memory to a spec)

  5. Fact-check. Spot-check the top-N memories by severity × staleness (days since updated). For each:

    • Check the cited source (code path, doc, ADR) against the current tree
    • If correct, no action
    • If drifted, correct with doctrine memory edit <REF>
    • If obsolete, supersede (memory status <REF> superseded --by <SUCCESSOR>) or quarantine (memory status <REF> quarantined)
    • If unverifiable, lower trust (memory edit <REF> --trust low)

  6. Report. Produce a brief handoff summary:

    • Actions taken (validated N, pruned M, linked L, fact-checked F)
    • Findings deferred to backlog (with ids if filed)
    • Items flagged for human attention (high-severity, ambiguous, needs decision)
    • Write this to notes.md or the active handover so the next agent doesn't re-do the same checks.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.