agentsclimarketplace

Update project memory

Skill DeltaCorvi/Claude-Skills-for-Smart-Crows/update-project-memory

A collection of reusable skills for Claude and other LLMs. Skills can be imported into Claude and other LLMs that use projects, or copy/pasted into any chatbot window.

Install
npx -y skills add DeltaCorvi/Claude-Skills-for-Smart-Crows --skill update-project-memory

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

2 things to look at

  • 26 days oldThe repository was created 26 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
  • 0 stars0 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

Reconcile saved memory against a project's actual working files, report discrepancies, and update memory after approval. Use when the user asks to review, audit, sync, reconcile, or update memory for a project; to check whether memory matches the files; or says memory is stale or out of date.

SKILL.md

3.7 KB, 753 tokens by cl100k_base, as published. Nobody here has run it

Sync Project Memory

Reconcile what is recorded in memory against the real state of a project's files, surface the differences, and update memory only after the user approves.

When to use

Trigger when the user wants memory checked against reality for an ongoing project: "review memory for this project", "is my memory up to date", "sync memory with the files", "audit what you remember about X", "memory is stale".

Do NOT trigger for general memory writes that have nothing to compare against (just saving a new fact), or for editing files unrelated to memory.

Inputs

  • The project (named by the user, or the active workspace folder if obvious).
  • The memory directory and its MEMORY.md index (paths are in system context).

If the project or its working folder is ambiguous, ask which one before proceeding. Do not guess.

Procedure

  1. Read memory. Read MEMORY.md and every memory file that relates to the project. Note what each one currently claims.

  2. Read reality. Enumerate the project's working folder (recursive listing), then read the key content files (outlines, drafts, notes, profiles, anything that states facts about the project). Do not rely on filenames alone; open the files.

  3. Diff. Compare memory against the files and sort every finding into one of three buckets:

    • Wrong: memory says something the files contradict.
    • Missing: the files establish something memory does not record.
    • Still accurate: memory matches; call this out so it is kept, not re-litigated.

  4. Report and hold. Present the three buckets plainly, then give concrete recommendations (which files to edit, add, or delete). Ask a clarifying question ONLY where there is genuine ambiguity that changes what you would write (for example an identity or name mismatch). Then stop and wait for approval. Do not write to memory in this step.

  5. Apply, on approval. Once approved:

    • Edit existing memory files in place with targeted changes. Do not rebuild or regenerate a file from scratch.
    • Add new memory files following the memory schema, choosing the right type (user, feedback, project, reference).
    • Cross-link related memories with [[name]].
    • Update MEMORY.md: one line per memory, no memory content in the index.
    • Convert relative dates to absolute. Confirm day-of-week or holiday constraints before committing a chosen date.

Judgment rules

  • Ask before assuming. When the files and memory disagree on something identifying (who the user is, which project this is, what a status means), ask rather than silently picking one. A wrong silent guess is the main failure mode this skill exists to prevent.
  • Targeted edits only. Change the specific fields that are wrong or missing. Leave accurate content untouched.
  • Do not duplicate the repo. Skip facts already recorded in the files themselves, code, or git history. Memory is for what is not derivable from the project's own materials.
  • Respect sensitive-info rules. Do not persist protected attributes, gov IDs, financial account numbers, health details, home addresses, or secrets unless the user explicitly asks.

Output

A short report (the three buckets plus recommendations) on review, and on approval a confirmation of exactly which memory files changed and how.

Gives 0 of the 12 instructions most memory context skills give in 753 tokens

Counted across 674 of the 847 authors here whose files we hold, read 2026-08-06

  • inform the user when setup is completein 21 of 674, across 6 files
  • confirm the draft with the user before writingin 21 of 674, across 6 files
  • update the agent skills block in place if it existsin 21 of 674, across 6 files
  • present findings to the userin 20 of 674, across 5 files
  • write the three docs files from seed templatesin 20 of 674, across 5 files
  • ask the user about each decision one at a timein 19 of 674, across 4 files
  • edit CLAUDE.md if it existsin 18 of 674, across 3 files
  • explore current repo statein 18 of 674, across 3 files
  • do not overwrite user edits to surrounding sectionsin 18 of 674, across 3 files
  • back up the original file before overwritingin 16 of 674, across 8 files
  • keep the memory index under 200 linesin 15 of 674
  • Provide actionable steps and verificationin 13 of 674, across 2 files

Said here and by no other author read

  • enumerate project working folder recursively
  • open content files instead of relying on filenames
  • sort findings into wrong missing or accurate buckets
  • give concrete file edit add or delete recommendations
  • edit memory files in place with targeted changes
  • add new files following memory schema

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

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.