share-context
Skill72 starsChanged 2 days agoApache-2.0View on GitHub
Shares Google Cloud operating context the user or their team already have — environments, projects, services and regions, product preferences, best practices, architecture guidance, runbooks — found in agent instruction files, local memory, skills and workflows, and the repository's GCP docs, with the Cloud Operator Agent's memory after the user reviews the list. Use when the user runs /cloud-operator:share-context, accepts the cloud-operator or setup skill's offer to share their existing Google Cloud context, or asks to share, import or save that context or a GCP fact for Cloud Operator. Not for ordinary GCP requests.
In this repository2 other files it gives its agents.
- cloud-operatorSkill
- setupSkill
The file
We split this skill into the rules it gives an agent
7 rules. About 0 are common, 6 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈6
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 26
If text follows the command, the user asks to save one GCP fact, or they pick their version after a conflict (Send, step 6), skip the …
- Distinctive: line 56
Put any other note about how they operate in Google Cloud in the closest group.
- Distinctive: line 67
Only "shared on" and "declined on" stop the cloud-operator and setup skills' offers.
- Very specific: line 72
Use one new operator session for the whole share: no contextId at first, then each reply's context_id.
- Distinctive: line 74
Write in plain language and the user's voice, without tags or structured formats, which the safety filter can reject.
- Distinctive: line 90
If a call fails with a permission error while earlier calls worked, retry once in a new session with everything sent so far, in the same …
- Distinctive: line 97
If its reply doesn't say the conversation is being saved (for example, "nothing new to save"), the save isn't confirmed and memory may be …
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Share existing Google Cloud context with Cloud Operator
Scan agent instruction files, local memory, skills and workflows, and the
repository's GCP docs for context the user or their team already have about
operating in Google Cloud, show it to the user, and copy what they approve into
the Cloud Operator Agent's memory. Use the cloud-operator skill's project,
agentId, secrets rules and usage attribution, including asking before using
their default gcloud project. Save the project once a call works.
One fact
If text follows the command, the user asks to save one GCP fact, or they pick
their version after a conflict (Send, step 6), skip the review. Follow
Send, /reflect included, with the message "Please remember this about
how I work in Google Cloud for future sessions. It's current and replaces
anything you have that conflicts: {fact}". Then confirm in one line.
Existing Google Cloud context
- Ask first, unless the user just agreed to an offer that named the
sources. Name them: agent instruction files (
CLAUDE.md,AGENTS.md,GEMINI.md,.claude/rules), local memory, skills and workflows written for this user, and GCP docs in this repository. Say that the share takes only notes about how they operate in Google Cloud, takes no secrets, and shows the list before sending. If local memory notes a confirmed earlier share ("shared on"), say when and offer only what's new. Check local memory only for an earlier "shared on" note; read nothing else until they agree. If they say no, note "declined" (step 4) and stop. Collect in three groups, copying what the sources say (names, dates, reasons) rather than searching past transcripts:
- Facts and preferences: environments (test, staging, prod) and their projects, other projects of interest, the services they manage, regions, networks, where IaC and state live, naming, owners, product preferences, best practices, architecture guidance (standard patterns, approved services, network/landing-zone conventions) and other standing choices.
- What happened: incidents and fixes, migrations, decisions and failed attempts, each with its date and outcome.
- How things are done: deploys, rollbacks, failovers, incident response, quota or access requests, each as a short summary naming its resources, and where its runbook is.
Put any other note about how they operate in Google Cloud in the closest group. Leave out secrets, personal and non-GCP notes, and live state such as metrics or bills. Mark notes that apply to one repository only. When local sources disagree, keep the newer one (by its date) or ask; a conflict with what the operator already has is settled after the save (Send, step 6). If nothing qualifies, say so and stop.
- Show the list, one line per note with its source, and ask: all, some or none?
- Send the picks, then note in local memory "Cloud Operator context:
shared on {date}", "declined on {date}" (no or none) or, if
/reflectdoesn't confirm, "not confirmed on {date}". Note nothing if a call failed. Only "shared on" and "declined on" stop the cloud-operator and setup skills' offers.
Send
- Use one new operator session for the whole share: no
contextIdat first, then each reply'scontext_id. Write in plain language and the user's voice, without tags or structured formats, which the safety filter can reject. Describe procedures as what the user does. Leave out empty groups, and above about 30 notes send one group per message:
These are notes about how I and my team operate in Google Cloud. Please remember them for future sessions and skip what you already have. If one conflicts with what you know, keep yours and tell me which, with your version and when you learned it. Reply with one line per note. Facts and preferences: - ... What happened: - ... How we do things: - ...- If a call fails with a permission error while earlier calls worked, retry once in a new session with everything sent so far, in the same wording, so the operator skips notes it already saved. If that fails, tell the user the operator refused; mention IAM only if other calls fail too.
- Send
/reflectalone, once, after the last message, without waiting on anything such as a conflict. The operator may save during the turn too;/reflectsaves everything now. - If its reply doesn't say the conversation is being saved (for example, "nothing new to save"), the save isn't confirmed and memory may be off. Tell the user they can retry later with the command.
- Otherwise, report briefly what was shared, skipped or flagged. If the operator kept its own version of a note, show the user both versions, each with its source and date if known: theirs from the list, the operator's from its reply. Ask once, for all conflicts, which version is current. Where they pick theirs, send it with One fact, all in one message; it replaces the operator's version. Where they pick the operator's or don't answer, leave it. If the operator still keeps its own version after that, tell the user and stop.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

