Conversation
The three self-hosted sandbox demos (docker, docker-memory, archil) now declare their agent, environment, and memory store as files that `ant apply` reconciles, instead of piping YAML into `ant beta:* create` from a hand-written agents/setup.sh. `ant apply` shipped in ant 1.30.0 (2026-09-03) and the docs page is public: https://platform.claude.com/docs/en/cli-sdks-libraries/cli/apply Per demo: - agents/<name>.md holds the agent (frontmatter is the request body, the prose is the system prompt), environments/self-hosted.yaml the self-hosted environment, and in docker-memory/ memory_stores/user-preferences.yaml the store. Same request bodies as the old YAML; agents/setup.sh and the agents/<name>/ directories are gone. - `ant apply .` creates everything and records the IDs in claude-lock.json. start.sh reads the environment ID from there with jq (an exported ANTHROPIC_ENVIRONMENT_ID still wins), archil/fanout.sh reads the agent and environment IDs from there, and the README session commands read them the same way. .env now carries only secrets: the environment key, optionally an API key, and archil's credentials. - claude-lock.json is gitignored here because every reader applies to their own workspace; the READMEs say to commit it in a real project. - READMEs, the shared README, and the CLAUDE.md runbook describe the apply flow (including --yes for Claude Code, which has no TTY). The ant pin in docker/Dockerfile and archil/sandboxes.py moves to 1.30.0 so one version covers host and sandbox. Verified against the API with ant 1.30.0: applied each demo (2 + 3 + 2 resources created), confirmed the server-side agent bodies match the old YAML, edited an agent and reapplied (version 2), reran as a no-op, ran the README `sessions create` commands and fanout.sh with lockfile-read IDs, checked start.sh resolves the environment from the lockfile, then deleted the sessions and pruned (archived) every test resource. Co-Authored-By: Claude <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AKwbEzQXeNMu5gXRTxbsag
… ID guards Three findings from the pre-review pass on the ant apply migration: - start.sh (all three) and fanout.sh only read the claude-lock.json beside them, but `ant apply` run from another directory writes the lockfile there and later runs walk up and adopt it. The "no environment ID" message now says to run apply from the demo directory so the lockfile lands beside the scripts, and the CLAUDE.md debugging table has a row for the "ant apply says up to date but start.sh finds nothing" loop. - docker/ and docker-memory/ are already published with the setup.sh flow, whose IDs live in .env. start.sh honors CLAUDE_ENVIRONMENT_ID again so a returning user keeps their environment and key instead of applying duplicates, and CLAUDE.md says what their options are. - fanout.sh reads IDs with `// empty` and a `:?` guard, so a partially applied lockfile fails with "run ant apply again" instead of sending the literal string null as an agent ID. Co-Authored-By: Claude <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AKwbEzQXeNMu5gXRTxbsag
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Description
ant applyshipped inant1.30.0 (2026-09-03) and its docs page is live: Manage resources as code with ant apply. This starts moving the Managed Agents quickstarts onto it, beginning with the three self-hosted sandbox demos (docker/,docker-memory/,archil/), which all provisioned their resources by piping YAML intoant beta:* createfrom a hand-writtenagents/setup.sh.Stacked on #460 (the
archil/quickstart) so the three demos and their shared README andCLAUDE.mdrunbook move together. GitHub retargets this tomainwhen #460 merges.Per demo:
agents/<name>.mdis the agent: frontmatter is the request body, the prose is the system prompt.environments/self-hosted.yamlis the self-hosted environment, anddocker-memory/addsmemory_stores/user-preferences.yaml. Same request bodies as the old YAML.agents/setup.shand theagents/<name>/directories are gone.ant apply .creates everything and records the IDs inclaude-lock.json.start.shreads the environment ID from there withjq(an exportedANTHROPIC_ENVIRONMENT_IDstill wins, for a sandbox host without the lockfile),archil/fanout.shreads the agent and environment IDs from there, and the READMEsessions createcommands read them the same way..envnow carries only secrets: the environment key, optionally an API key, archil's credentials.claude-lock.jsonis gitignored in each demo because every reader applies to their own workspace. The READMEs say to commit it in a real project.CLAUDE.mdrunbook describe the apply flow, includingant apply --yes .for Claude Code (no TTY for the confirmation prompt) and the origin-mismatch error. Theantpin indocker/Dockerfile(1.23.0) andarchil/sandboxes.py(1.28.0) moves to 1.30.0 so one version covers host and sandbox.Not in this PR:
assistant-ui/,chat-sdk/,copilot-kit-ag-ui/still provision from TypeScript setup scripts here (and fromagents/setup.shYAML on the private repo'smain, which never published). They are the follow-up, same pattern.Quickstart
Type of Change
Testing
Against the API with
ant1.30.0 (linux amd64 release binary), in a test org:ant apply --dry-run -v .in each demo: plans 2, 3, and 2 creates with the same bodies the old YAML sent (system,tools,metadata,config: {type: self_hosted});README.md, scripts, andDockerfileare skipped by the walk.ant apply --yes .in each demo: all 7 resources created,claude-lock.jsonwritten with./agents/...,./environments/...,./memory_stores/...keys.ant beta:agents retrieveon each agent matches the file (archil's 3.8 KB system prompt intact).docker-memory/agents/memory-demo.md: plan shows~ update, agent goes to version 2, lockfile records it. Immediate rerun:Everything is up to date.ant beta:sessions createcommands fordocker/anddocker-memory/(with--resource {type: memory_store, ...}) run verbatim with thejq-read IDs; the session comes up on agent version 2 with the store attached.archil/fanout.sh "Tesla"creates its session from the lockfile IDs.start.shin all three resolves the environment fromclaude-lock.jsonand stops at the expected next check (set ANTHROPIC_ENVIRONMENT_KEY ... for env_..., or the archil SDK check).ant apply --prune --yesarchived every test resource.Not run end to end: a worker actually serving a session (needs an environment key minted in the Console), the archil data load, and the
docker buildwith the 1.30.0 pin (the release asset URL pattern is unchanged andbeta:worker runkeeps--workdir/--max-idle/--log-formatin 1.30.0).Additional Notes
--no-managed-metadataflag and says apply addsmanaged_bymetadata, but 1.30.0 has neither (ant apply --help, and the created agents carry only the file's ownmetadata). Nothing in this PR depends on it.🤖 Generated with Claude Code
Requested by CJ · Claude session #01AKwbEz