Drop markdown files in a folder, number them, and run. Each file is one task sent to an AI agent. When you need loops, retries, or conditional logic — add an .ash script. Start simple. Grow as needed.
Ash runs a folder of markdown files through AI agents in numbered order — no scripting required. Each file is a task. Frontmatter sets the agent and model. Output flows between steps.
tasks/
├── 01-research.md
├── 02-implement.md
├── 03-review.ash
└── 04-deploy.md---
agent: opencode
model: sonnet
---
# Research the login system
Trace all files in the login flow.
Document responsibilities, inputs,
and test coverage per file.#!opencode:1.0
RESULT = $(npm test 2>&1)
if $? != 0 {
do "Tests failed. Fix errors
and recompile.
Test output:
${RESULT}" with opencode
}
Markdown and .ash files sort together by numeric prefix. Choose whichever fits each step.
You never face more than you need. Start at level 1, add one concept at a time.
| Level | What you use | What you get |
|---|---|---|
| 1 | Folder of .md files |
Sequential AI task execution |
| 2 | YAML frontmatter | Per-task agent, model, on_fail |
| 3 | .ash files in the tree |
Logic within a single step |
| 4 | ${stdout}, $? |
State passing between tasks |
| 5 | Standalone .ash scripts |
Full orchestration: loops, retry, parallelism |
When the workflow itself needs programming — iterate over files, retry on failure, branch on results — standalone .ash scripts give you the full pipeline.
MSG = "hello"
ITEMS = ["a", "b", "c"]
FIRST = ITEMS[0]
print "count: ${len(ITEMS)}"
working_dir = $(pwd)do "Review this code" with opencode
try {
do "Fix the bug" with fixer
} fail {
do "Retry: ${stderr}"
} upto 3for FILE in FILES {
do "Review ${FILE}"
if $? != 0 { exit 1 }
}
exec npm test
if $? == 0 {
print "all good"
}fn review(path) {
do "Review ${path}" with opencode
}
include "helpers.ash"
review("src/main.rs")AI agents handle ambiguity — understanding intent, making judgments. They're bad at sequencing, branching, and retrying. Ash handles the structure. Agents handle the content.
Tasks run in numbered order every time. No "what should I do next?" — the script decides. The agent executes.
The agent decides how to do its step. Ash decides whether the step happens, how many times, and what comes next.
A weekly report, a recurring audit, a batch job — same process every time, even when the content of each step varies.
Pre-determined control flow runs at CPU speed. AI time is spent only where it adds value — on the content of each step.
Review content, audit compliance, publish updates — all live in files you commit, version, share, and re-run. Not buried in chat threads.
OpenCode, Claude Code, Aider, or a custom tool. Define in a config file. Swap providers without touching your workflows.
Write and run Ash scripts in your browser, or use the REPL for interactive experimentation. The built-in js-echo agent echoes your prompt back.