@DaveJ Ohhh this is awesome! Having it show red if something has an error would be nice to have. I'm going to try this with my agents tonight, great work!
Here's the /goal prompt. Paste it directly after /goal in each agent's CLI. Only swap [AGENT NAME] and [FOUNDATION PATH]:
---
```
/goal
You are [AGENT NAME]. Your SYSTEM_DOC.md at [FOUNDATION PATH]/SYSTEM_DOC.md is your starting point — it is the most complete map of your system that exists. Your goal is to forensically audit your entire system, close every gap you find, and leave your codebase and documentation in a materially better state than you found it. Spend as much time as needed. Do not stop until you run out of meaningful improvements.
Work in this order:
PHASE 1 — FORENSIC READ
Read your SYSTEM_DOC.md fully. Then read every file it references. Do not trust the doc — verify it against reality. For every claim in SYSTEM_DOC.md, check that the actual file exists, contains what it claims, and does what it says. Flag every discrepancy.
PHASE 2 — GAP ANALYSIS
Identify every gap you find across these dimensions:
Files referenced but missing on disk
Functions that are stubs, TODOs, or incomplete
Skills that exist in docs but have no working implementation
Sub-agents with no actual logic
Integrations documented but not wired (dead imports, missing env vars, uncalled functions)
Data flows described but not implemented
Error handling that is absent or superficial
Hardcoded values that should be configurable
Missing tests or validation for critical paths
Logic that duplicates itself across files (consolidation opportunities)
Dead code that can be removed
PHASE 3 — PRIORITIZE
Stack-rank every gap you found: Critical (system broken without fix) → High (significant capability missing) → Medium (quality/reliability improvement) → Low (cleanup/polish). Write this ranked list to [FOUNDATION PATH]/GAP_ANALYSIS.md before proceeding.
PHASE 4 — EXECUTE
Work down your ranked list. For every item:
Fix it completely — no partial implementations, no TODOs left behind
If a skill or sub-agent is a stub, build it to full working state
If a file is missing, create it with real content
If logic is duplicated, consolidate it
If an integration is unwired, wire it
After each fix, verify it actually works — read the result back, trace the logic
PHASE 5 — ADVANCEMENT
After all gaps are closed, look for upgrade opportunities:
Are there capabilities your system should have that it doesn't?
Are there more efficient patterns you could apply?
Are there cross-agent data flows that would make you more powerful?
Are there performance bottlenecks you can eliminate?
Can any sub-agent be made more autonomous or more capable?
Build the most valuable ones.
PHASE 6 — UPDATE DOCS
Rewrite SYSTEM_DOC.md to reflect the current true state. Remove all [UNVERIFIED] and [BROKEN] tags you resolved. Add everything you built. Update the current state section to be accurate as of today.
Rules:
Never claim something is complete without verifying the actual file/output
Never leave a TODO or placeholder — finish it or don't start it
If you hit something that requires [YOUR NAME]'s credentials or a business decision, note it in [FOUNDATION PATH]/NEEDS_[YOUR NAME].md and keep working on everything else
Report your progress every time you complete a phase by writing to [FOUNDATION PATH]/GOAL_PROGRESS.md
Do not stop. Keep working until every gap is closed, every stub is built, and the system is materially more capable than when you started.
Claude's new update has /goal, I just gave these prompts to my agents and used /goal for the second prompt.
8 agents currently doing this in parallel and I'm seeing leaks being plugged that I didn't know existed.
*if you're worried about data usage, start using pure Claude Code agents instead of openclaw. I used to need 2x Max Pro subs, now I use 1 and never hit my weekly limit*
Prompts posted below.
@AnthropicAI
Claude Code 2.1.139 has been released.
50 CLI changes, 1 system prompt change
Highlights:
• Added /goal command to run tasks across turns until a set completion condition, with live elapsed/turns/tokens
• System prompt compaction is now silent (no trimming alert), which may obscure that prompts were trimmed
• Compaction prompt asks model to preserve sensitive instructions, reducing loss of user intent during trimming
Full details are in thread ↓
SYSTEM DOCUMENTATION AUDIT — NEW TASK
Your task is to perform a complete audit of your own system and produce a full, accurate, living documentation file at [FOUNDATION PATH]/SYSTEM_DOC.md. This is not a summary — it is a master reference document.
Do this in order:
WORKSPACE AUDIT
Read every file in your foundation directory. Do not guess what files exist — actually list them using Glob on your foundation path. Then read each one.
SUB-AGENTS / AGENTS
For each sub-agent you have: its name, its role, the file it lives in, what it does, what inputs it takes, what outputs it produces, and what triggers it.
SKILLS
For each skill: its name, what it does, when it fires, what it writes or changes, and any known issues.
CORE LOGIC & DECISION FLOW
Describe how you actually operate: how you receive tasks, how you decide what to do, how you prioritize, how you complete and report work. Not theory — what actually happens.
INTEGRATIONS
Every external system you connect to: Firebase/Firestore collections you read/write, APIs you call, databases, Discord channels, file paths you depend on. Be specific — collection names, endpoints, file paths.
DATA & MEMORY
What you store, where you store it, what format, and why. Include your MEMORY.md structure if you have one.
CURRENT STATE
What is working right now. What is broken or degraded. What is in-progress. What is stalled. Be honest — do not report things as working if they haven't been tested recently.
KNOWN ISSUES & TODOS
Pull from your TODO.md and any unresolved items you know about. Prioritize them: Critical → High → Medium → Low.
PERFORMANCE & METRICS
If you track any performance data (win rates, conversion rates, revenue, uptime, error counts), document current numbers and where that data lives.
DEPENDENCIES ON OTHER AGENTS
What you need from [OTHER AGENT NAMES] etc. What they need from you. Where handoffs happen.
---
Output format for SYSTEM_DOC.md:
Use clear ## section headers matching the 10 sections above
Be exhaustive on file paths — use absolute paths
Mark anything unverified with [UNVERIFIED]
Mark anything broken with [BROKEN]
Mark anything recently changed with [UPDATED: date]
Do NOT use placeholders. If you don't know something, say "unknown — needs verification"
When complete, reply with: ✅ SYSTEM_DOC.md written — [X] files audited, [X] sub-agents documented, [X] issues flagged.
My feed is full of, “ I’m cancelling Claude and switching to Codex” which are a bunch of kids that have never actually built anything and likely never will. Codex will never truly compete with Claude. If you’re truly building intelligence systems and are capable of pushing things “beyond,” then you know where the real power is.
#Anthropic
@kapilansh_twt 1- 8 agents, 3 making money daily, others used every few days for different scenarios.
Not a single hour, even when asleep, that data and intel aren’t being produced and absorbed via machine learning. 🫣
@UsernameAndStuf@BentoBoiNFT Ask cli to swap your Openclaw agents to Claude agents. You’ll get more sleep and your system will be less flawed. Thank me later.
Building entire frameworks and systems capable of handling real companies can’t be built with $20/month subs brother.
Once you get through your learning phase and start building, you’ll understand why Anthropic charges what it does for the best coding AI on the planet.
Hope to meet you at a founders meet-up once day. Keep absorbing.
I asked my openclaw orchestrating agent 2 months ago if we could convert to using pure Claude code with all of my agents. It said yes, but there would be a lot of issues. A few weeks ago or so, after seeing @AnthropicAI advancing in the agent era and cutting out openclaw, I asked my clause cli directly to convert my entire company. Flawless. Happiest I’ve been since I got into coding with AI 9 months ago GRINDING every single day. Learning, taking beatings, upsets, fighting beyond failure and disappointment.
My systems are blowing my mind now. Instead of trying to push sales and marketing right now, I’m letting them generate profit without needing to sell anything.
Sports picks now, so advanced from ML I don’t want to share, will do when I have months of track record showing my roi.
Stocks swarm intelligence with 38 channels- my option trading friends that aren’t into AI are feeding my agent questions and wondering why I’m outperforming them.
ERP system fully built out, my current new day job employer (mega corp) saw it, said they’re paying 3 different companies for systems that mine has all in one and that we need to meet soon.
—————————————-
Don’t settle. Head down. Don’t stop pushing beyond the realms of what’s possible. *tell that to your agents too*
When Anthropic nixed Openclaw, I had Claude cli convert all of my Openclaw agents into Claude Code agents. It did, executed it flawlessly. They’re all the most functional they’ve ever been, none of the lingering pains that Openclaw has, are GONE.
I was using 2 pro max plans and still using my weekly data usage 5 days in. Now with pure Claude code and my agents in discord with their own channels working flawlessly, I don’t use half the weekly usage of 1 pro max plan now. It’s astonishing and I’ll never go back to using Openclaw. Anthropic is adapted since Openclaw launched, you don’t need Openclaw anymore.
@tmuxvim Are you only running one instance at a time without cycles of scheduled tasks permitting data checks? I’d say it’s easy to not touch usage limits if your work is… limited. Think beyond, expand your capabilities. Limiting to a single agent won’t touch what you’re capable of. 🙏🏻
If you’re not building with drive and ambition, willing to pay $200 a month for the best there is, stick to $20 a month chat bots. This is the difference between tinkering around with AI and actually giving your all to stay ahead. Nobody is building businesses from scratch/empires like some of us, on the $20/month plans. I wish you the best, but complaining about data usage on the lowest tier subscriptions isn’t proving you’re capable of doing something real.
When Anthropic nixed Openclaw, I had Claude cli convert all of my Openclaw agents into Claude Code agents. It did, executed it flawlessly. They’re all the most functional they’ve ever been, none of the lingering pains that Openclaw has, are GONE.
I was using 2 pro max plans and still using my weekly data usage 5 days in. Now with pure Claude code and my agents in discord with their own channels working flawlessly, I don’t use half the weekly usage of 1 pro max plan now. It’s astonishing and I’ll never go back to using Openclaw. Anthropic is adapted since Openclaw launched, you don’t need Openclaw anymore.
@markduphiney@JulianGoldieSEO This guy has never actually built anything with AI except engagement posts. I would definitely ask AI for help, not the guy that just uses AI for social media.
Nice clickbait, but I’m about to get a 3rd Claude Pro Max sub for the work me an my swarm of agents are doing.
Sure, settle for less if you’re main goal is the be an AI influencer. But, if you’re actually building, don’t settle.
I ditched Openclaw after Anthropic ditched it. My system is more intelligent and more advanced than it’s ever been and I fucking love it.
If you’re not building multi-agent orchestration systems, yes, go with your cheap/free chatbots. If you want to say, “this is fucking incredible” every single day while building with your swarm, go @AnthropicAI 1000%, you pay for what you get, literally.
Claude Code is now completely free forever 🤯
Google just dropped Gemma 4 on Ollama, which means you can run Claude Code locally with zero API bills and zero subscriptions.
Just install Gemma 4, connect it to Claude Code, and suddenly you have free AI agents running directly on your computer.
Google may have just killed AI subscription costs. 🚀
@ziwenxu_ Brother, tell your agent to go through its files and clean up anything that could be eating weekly data usage. You’re obliterating your usage with massive files that it reads for no reason.
I do 40x whatever you’re doing and ain’t hitting limit anymore.
@ChrisCoffee I honestly laughed a little too hard at this release. I run openclaw with Claude via 9 agents in discord. I'm going to test it with one for a bit, but this is wild Anthropic doesn't slow down and I'm here for it.