orchestrator-loop
Skill2 starsChanged 9 months agoMITView on GitHub
Orchestration pipeline for development tasks. The orchestrator delegates everything and never writes code directly.
In this repository27 other files it gives its agents.
- Agent BrowserSkill
- agent-governanceSkill
- agent-system-learning-loopSkill
- agent-system-live-quality-opsSkill
- agent-system-memory-proactivity-opsSkill
- agent-system-output-hygiene-opsSkill
- agent-system-stability-opsSkill
- auto-updaterSkill
- deep-researchSkill
- domaindetailsSkill
- hnSkill
- humanizerSkill
- memory-hygieneSkill
- model-usageSkill
- nano-pdfSkill
- news-aggregator-skillSkill
- prompt-guardSkill
- security-reviewSkill
- self-improvementSkill
- structured-brainstormingSkill
- subagent-executionSkill
- summarizeSkill
- systematic-debuggingSkill
- tavilySkill
- tdd-workflowSkill
- topic-monitorSkill
- verification-gateSkill
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 5 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
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 21
If context is already clear --> SKIP (e.g., user provided file paths and explicit instructions)
- Very specific: line 22
If context is missing --> delegate via bridge.sh --research "..." or a research subagent
- Distinctive: line 29
HARD-GATE: No implementation without approval.
- Distinctive: line 41
Check: spec compliance + code quality + tests
- Distinctive: line 42
If it does not pass after 3 iterations --> escalate to the user.
- Distinctive: line 56
Use the shortcut flow when ALL of the following are true:
≈ “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
Orchestrator Loop
The orchestrator DOES NOT WRITE CODE. This pipeline defines how the orchestrator delegates tasks.
Pipeline
RESEARCH --> PLAN --> IMPLEMENT --> REVIEW --> REPORT
| | | | |
delegate orchestr. delegate delegate orchestr.
(bridge.sh) (approve) (spawn) (spawn) (report)Step 1 -- RESEARCH (Gather Context)
- If context is already clear --> SKIP (e.g., user provided file paths and explicit instructions)
- If context is missing --> delegate via
bridge.sh --research "..."or a research subagent - Output: Summary of current state + relevant files + risks
Step 2 -- PLAN (Design + Approval)
- Large work (multi-file, architectural decisions): invoke the
structured-brainstormingskill --> get user approval - Small work (single file, clear instructions): orchestrator writes a short summary --> get user approval
- HARD-GATE: No implementation without approval. Do not proceed to Step 3 until the user says "go".
Step 3 -- IMPLEMENT (Delegate)
- Invoke the
subagent-executionskill --> spawn subagents for each task - The orchestrator only WATCHES: answers subagent questions, redirects if stuck
- The orchestrator DOES NOT WRITE: creating/editing files directly is forbidden
- Alternative for simple tasks: delegate via
bridge.sh --code "..."to a coding subagent
Step 4 -- REVIEW (Verify)
- Invoke the
verification-gateskill - Check: spec compliance + code quality + tests
- If it does not pass after 3 iterations --> escalate to the user; do NOT try to fix it yourself
Step 5 -- REPORT (Present to User)
The orchestrator reports:
- What was done (changed files + summary)
- What was skipped and why
- What requires human review
- Known limitations, if any
Shortcut Flow
Use the shortcut flow when ALL of the following are true:
- Single file change
- Less than 20 lines changed
- Low risk (does not touch a working system)
- Clear instructions (what to do is unambiguous)
Shortcut pipeline:
- Orchestrator writes a short summary --> gets user approval
- Delegates to a single coding subagent via
bridge.sh --code "..." - Verifies the result (read file + run)
- Reports
Red Flags
| If the orchestrator is thinking... | Reality |
| "This is simple, I will do it myself" | NO. Simplicity does not change the rules. Delegate. |
| "Just one line" | Size does not matter. Writing directly is forbidden. |
| "Spawning a subagent is slow" | Doing it right > doing it fast. |
| "The pipeline is overkill" | Use the shortcut flow, but do NOT skip the pipeline. |
| "I already know the answer" | Knowing is not writing. Delegate. |
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.

