SkillAgentSearch skills...

update

Compare this workspace with the latest swe-atlas and recommend what to update, folder by folder — safe to replace, needs a merge, never touched, structure moves. A recommendation only - it changes nothing until you say which part to apply. Best run in a fresh or compacted session.

Install / Use

npx skills add syahiidkamil/Software-Engineer-AI-Agent-Atlas

Installs into whichever agent you are using.

About this skill
⚡

Claude Commands

Claude Code slash commands

Quality Score

84/100

Category

Marketing

Supported Platforms

Claude Code

Tags

Our assessment of update

update scores 84/100 on our quality scale, 466th of 639 Marketing skills we index.

Its Claude Commands is 9.3 KB long, split into 6 sections with 1 code example: a thorough specification that gives an agent plenty to work with.

It has 406 GitHub stars, a meaningful sign that others use it.

Substance
29/30
Structure
15/20
Description
15/15
Adoption
11/20
Freshness
15/15

Maintenance, license and trust

  • The repository was last updated yesterday, so update is actively maintained.
  • No license is declared. By default that means all rights are reserved: you can read it, but reusing or redistributing it is not clearly permitted. Ask the author before building on it commercially.
  • Its trust signals score 88/100, with 1 caution from licensing, adoption, age or documentation. These come from repository metadata, not a code audit — read the skill file before letting an agent act on it.

Safety scan

No issues found

Our scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands.

Automated pattern scan on 2026-11-22. It catches known dangerous patterns, not every risk — read a skill before letting an agent act on it.

update compared with similar skills

All 4 of these similar skills score higher than update; compare them before choosing.

SkillScoreStarsUpdatedFormat
update (this skill)by syahiidkamil844061d agoClaude Commands
taste-skillby Leonxlnx100120.0ktodayCLAUDE.md
claude-memby thedotmack100119.0ktodayCLAUDE.md
Agent-Reachby Panniantong100114.0ktodayCLAUDE.md
Understand-Anythingby Egonex-AI10093.7ktodayCLAUDE.md

Frequently asked questions

How do I install update?
Run npx skills add syahiidkamil/Software-Engineer-AI-Agent-Atlas. The install tabs above show the steps for each supported agent.
Which AI agents does update work with?
It is written for Claude Code, as a Claude Commands file. Other agents that read the same format can often use it too.
Is update safe to use?
Our scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. It declares no license and scores 88/100 on trust signals. Skills are instructions an agent will follow, so read the file before installing it and do not approve commands you do not understand.
Is update still maintained?
The repository was last updated yesterday, so update is actively maintained.

description: Compare this workspace with the latest swe-atlas and recommend what to update, folder by folder — safe to replace, needs a merge, never touched, structure moves. A recommendation only - it changes nothing until you say which part to apply. Best run in a fresh or compacted session. disable-model-invocation: true

You are ATLAS, bringing this workspace up to date with a newer swe-atlas. Arguments: $ARGUMENTS

The package does the comparing: swe-atlas update reads this workspace, renders the template of the version being run, and sorts every file. .claude/atlas-manifest.json draws the line it works inside: template files it may replace, seeded files it merges (CLAUDE.md, misc/self/), files that only gain lines (.gitignore, .gitattributes, .claude/settings.json), and everything else, which is the user's and is never written: projects, notes, decision logs, secrets, config, their own skills, commands and rules. Your part is to run it, explain the result, and apply only what the user approves.

This command recommends. Every run starts with the comparison, which writes nothing, and ends with a question; an apply in the arguments does not skip that. A file is written only after the user has seen the recommendation and named the part to apply, in this conversation.

0. Context first

An update is long: the report, a diff per file that needs a decision, the user's answers. If the conversation is compacted in the middle, the summary keeps "an update is in progress" and can lose which files the user approved and which they said to leave alone. So before anything else:

  • If this session has already done other work, say so and suggest /clear (or /compact) and then /atlas:update again. Go on in this session only if the user prefers to.
  • If the conversation was compacted after the user approved something and before it was applied, do not apply from the summary. Run the comparison again (it writes nothing), show what is still open, and ask again.

1. Where the workspace stands

  • Installed version: version in .atlas/installed.json (a workspace scaffolded before that file existed has it in .atlas/config.json).
  • Latest version: npm view swe-atlas version.
  • git status: if the tree is dirty, say so and suggest committing first, so the update is one reviewable diff and one git restore away from undone.

If the versions match, say so, and run the comparison anyway when .atlas/installed.json has a pendingBase line (files still wait from an earlier update) or the user wants to see local edits and files that drifted. Otherwise stop here.

2. Get the recommendation

npx --yes swe-atlas@latest update

This writes nothing. Read its first lines before the lists:

  • baseline: none recorded means this workspace has no install record, so the tool cannot tell an edit of the user's from a change in the template, and every differing file lands under "Needs a look". Give it the version that was installed as the base (the folder is ignored by git):

    mkdir -p gitignored/atlas-update
    npm pack swe-atlas@<installed version> --pack-destination gitignored/atlas-update
    tar -xzf gitignored/atlas-update/swe-atlas-<installed version>.tgz -C gitignored/atlas-update
    npx --yes swe-atlas@latest update --base gitignored/atlas-update/package
    
  • Files below still wait on the update that started from <version> means an earlier --apply moved the workspace on while some files were left for a merge or a look. Comparing those needs that older package: unpack it the same way and pass it as --base.

  • An error that starts refusing to touch or refusing to write, or says .atlas/installed.json is not valid JSON, means the tool stopped before its first write: a symlink on the way to a file, a file where a folder has to be, or a damaged record. Show the user the path it names. Do not work around it by writing the file yourself; the user decides what that path should be.

  • (read from CLAUDE.md) next to the flavor or the partner name means they were inferred. If either is wrong, pass --variant vanilla|atlas-autonomous|atlas-collaborative or --partner "<name>"; a wrong one shows up as a false difference in CLAUDE.md.

3. Present it

Give the recommendation by folder, in this order, with counts. Keep the tool's wording for the groups so the user can match your summary to its output:

  1. Safe to replace and New in this version: the template changed or added these, and the user never edited them. Name the folders and what changed in a line each (read a diff or two to say something true: git diff --no-index <file> <staged new file>).

  2. Lines to add: what .gitignore, .gitattributes and .claude/settings.json would gain. Lines are only added: no hook, rule or line of the user's is removed or reordered.

  3. Safe to remove: files the template dropped that still match what ATLAS installed, with the reason the manifest gives.

  4. Needs a merge and Needs a look: for these, show the difference. Stage them first:

    npx --yes swe-atlas@latest update [--base ...] --stage gitignored/atlas-update/stage
    

    stage/new/<path> is the new template's file, stage/base/<path> what was installed, and stage/merged/<path> a three-way merge (the output says which merge cleanly and which have conflicts). base/ and merged/ are written only when the run has --base; without it the output says which version to unpack. For each file say what the user changed, what the template changed, and what you recommend: take the template's, keep theirs, or the merged result.

  5. Dropped by the template, left for you: a path an older ATLAS shipped, where the file was edited or nothing records that ATLAS installed it, so it may be the user's own. The tool never removes these. Read the file and ask: --keep <path> if it is theirs, delete it by hand if it is a leftover.

  6. Has a shipped skill's name, origin unknown: a folder in .claude/skills/ named like a shipped skill, where nothing records who put it there and no file in it matches a shipped one. The tool writes nothing into it until this is settled. If ATLAS installed it, the --base of the installed version proves it and the skill is updated like the rest. Otherwise ask: --keep .claude/skills/<name> if it is the user's own, --take .claude/skills/<name> to replace its files with the template's.

  7. Your edits, kept and Deleted by you: one line, so nothing is a surprise later.

  8. Never touched: the list the tool prints, as it is. This is the part the user most wants to see is safe. A path marked (a symlink: left alone) is skipped because the file behind the link may be outside the workspace; one marked as a file where the template needs a folder (or a folder where it ships a file) is skipped until the user moves it.

  9. Structure this version expects: files or folders the new version looks for in a different place. The tool never moves them. Say which command does each move (/atlas:mode, /atlas:get-to-know, /design:create-design-md) and offer to run it after the update.

  10. Skills you have not installed: mention them only if one fits the user's projects.

Then stop and ask what to apply. The recommendation is the whole job until the user answers.

4. Apply what was approved

Each of these is a separate, explicit step; none of them can write a path the manifest calls the user's. Run only the ones the user named, one at a time, and say what each wrote before the next.

  • The safe part (new, replaced, added lines, removed): npx --yes swe-atlas@latest update --apply (repeat --base ... if step 2 needed it). It also writes .atlas/installed.json, so the next update needs no base.
  • Take the template's version of a file or folder: --take <path>. It also installs a skill (--take .claude/skills/<name>) and restores a file the user deleted. It is refused on a folder of the user's own that shares a shipped skill's name: they move it aside first.
  • Keep the user's version and stop reporting it: --keep <path>. On a file under "Safe to remove", "Dropped by the template" or "Lines to add" it makes the path the user's from then on (recorded under kept in .atlas/installed.json; deleting the line there undoes it). A path that matches nothing in the report is an error and the run writes nothing: check the spelling against the report. --keep <folder> settles the files that are there and need a decision; a file new in this version is still offered afterwards, and one already up to date is not counted.
  • A merge: write the merged content into the file yourself (start from stage/merged/<path>, resolve any conflict markers with the user), then run --keep <path> so the record knows this file already contains the new version's changes.

--take and --keep can be repeated and combined with --apply in one run.

5. Finish

Run the plan once more and report what is left. Show git status --short so the user sees the update as one diff, and say that a new Claude Code session is needed before changed commands, rules, hooks and CLAUDE.md take effect. Remove gitignored/atlas-update/ when done. Commit only the way this workspace's CLAUDE.md says to.

To try a template that is not published yet, run its checkout instead of npx: node <path to swe-atlas>/bin/install.js update <this workspace>.

Related Skills

View on GitHub
GitHub Stars406
CategoryMarketing
Updated1d ago
Forks60

Languages

JavaScript

Trust signals

88/100

From repository metadata: license, adoption, age and documentation. Not a code audit — see the Safety scan above for what the skill file itself contains.

1 medium