Create docs on how to build your own coding agent harness #885

Activity

1. skrawcz

   added

   good first issueGood for newcomersGood for newcomers

   kind/documentationDocs, examples, notebooksDocs, examples, notebooks

   on Aug 15, 2026
2. github-actions

   added

   status/needs-triageNew issue, needs triageNew issue, needs triage

   on Aug 15, 2026
3. Agentic-JJ-Web3 commented on Aug 15, 2026

   

   Agentic-JJ-Web3

   on Aug 15, 2026

   More actions

   I'd like to take this on. I build LLM engineering curriculum and teach this
   material, so writing a walkthrough is in my wheelhouse.

   Before I start — what shape are you after? A conceptual piece on why a state
   machine fits an agent harness, or a working example under examples/ with a
   narrative doc alongside it? And is there an existing doc whose structure I
   should mirror?

   Happy to sketch an outline first for you to react to before I write anything.
4. skrawcz commented on Aug 15, 2026

   

   skrawcz

   on Aug 15, 2026

   Last edited by skrawcz

   ContributorAuthor

   More actions

   I think it's starting simple and then building things in chunks and steps.

   1. building the harness state machine loop. So what steps do we have in it? E.g. prompt creation, llm call, tool execution, etc.
   2. coming up with the basic tools needed. e.g. read file, write file, run bash.
   3. running it via code, writing tests, and seeing it work in the Burr UI.
   4. wrapping it in a CLI for a basic command line experience
   5. then showing what else could be done -- where would you add permissions? (add a new step before tool execution most likely) how would you enable "skills"? etc.

   Does that make sense and match the outline you were thinking of?
5. skrawcz

   assigned

   Agentic-JJ-Web3

   on Aug 15, 2026
6. Agentic-JJ-Web3 commented on Aug 16, 2026

   

   Agentic-JJ-Web3

   on Aug 16, 2026

   More actions

   1. LLM & CI Testing: Should the example use a mock LLM by default so it runs without an API key and works in CI, with a real client swappable in?
   2. Bash Security Sequencing: Step 2 includes a bash tool, but permissions don't arrive until step 5. Should the early version ship un-sandboxed deliberately so step 5 has something concrete to fix, or does it need guardrails from the start?
   3. Code Location: I plan to put this in examples/coding-agent/ with a docs page alongside. Is there an existing example you would like me to mirror for structure?
   4. Shipping Plan: I would like to ship this in three small PRs (steps 1–3, then the CLI, then permissions and extensions) to keep each review small and fast.
   5. Documentation Note: I noticed examples/hello-world-counter runs fine but crashes at app.visualize() unless graphviz is installed. I am happy to add a note to that example if it is useful!
7. github-actions commented on Sep 2, 2026

   

   github-actionsBot

   on Sep 2, 2026 – with GitHub Actions

   More actions

   @Agentic-JJ-Web3, this issue has been inactive for 17 days.
   Are you still working on it? Drop a comment to let us know.

   If there is no update within 7 days, the assignment will be removed
   so someone else can pick it up
   (per the assignment policy).
8. Agentic-JJ-Web3 commented on Sep 4, 2026

   

   Agentic-JJ-Web3

   on Sep 4, 2026

   More actions

   Still on it. PR #886 has been open since 16 Aug with steps 1–3 (harness loop, tools, Burr UI) — it covers most of the issue and hasn't been reviewed yet. @skrawcz, a few questions in the PR description are still open, mainly whether the scripted LLM client fallback is the approach you want. Happy to adjust once you've had a look.
9. github-actions commented on Sep 23, 2026

   

   github-actionsBot

   on Sep 23, 2026 – with GitHub Actions

   More actions

   @Agentic-JJ-Web3, this issue has been inactive for 19 days.
   Are you still working on it? Drop a comment to let us know.

   If there is no update within 7 days, the assignment will be removed
   so someone else can pick it up
   (per the assignment policy).
10. github-actions commented on Sep 30, 2026

    

    github-actionsBot

    on Sep 30, 2026 – with GitHub Actions

    More actions

    Removing assignment from @Agentic-JJ-Web3 after 26 days of inactivity
    (per the assignment policy).

    Feel free to comment if you want to pick this back up or if someone else wants to take it.
11. github-actions

    unassigned

    Agentic-JJ-Web3

    on Sep 30, 2026
12. github-actions

    added

    help wantedContributors wanted!Contributors wanted!

    on Sep 30, 2026