v0.6.1
loopcoder v0.6.1
v0.6.1 is the customer-ready bridge for the public 0.6 line. The latest public
release before this bridge was v0.5.4; customers should install or upgrade to
v0.6.1, not to a nonexistent public v0.6.0 release.
This release packages the 0.6 capabilities into an operable first-run customer
flow: model/depth discovery, Antigravity worker support, reporter output,
upgrade/migration compatibility, skill install, read-only doctor, local run
status, report querying, relay gates, loopreview, audit, state branch, lease,
and process watchdog behavior.
Install
Unix-like systems:
curl -fsSL https://raw.githubusercontent.com/jasonhnd/loopcoder/main/scripts/install.sh | sh -s -- --version 0.6.1
Windows PowerShell:
$env:LOOPCODER_VERSION = "0.6.1"
irm https://raw.githubusercontent.com/jasonhnd/loopcoder/main/scripts/install.ps1 | iex
The install scripts verify signed SHA256SUMS with cosign before trusting
release checksums.
Upgrade from v0.5.4 or older 0.5.x
loopcoder upgrade --version 0.6.1
loopcoder version
loopcoder skill install --repo .
loopcoder doctor --repo .
loopcoder upgrade --version 0.6.1 selects the machine-level binary and
refreshes the global bundled skill. Each project still needs
loopcoder skill install --repo <repo> to refresh project hook settings,
.loopcoder/conductor-workspace, and local .git/info/exclude protection for
.loopcoder/. Then run loopcoder doctor --repo <repo> to confirm readiness.
After v0.6.1 is selected, re-running loopcoder upgrade --version 0.6.1 should
report that the selected binary is already latest.
Customer quickstart
From a repository root:
loopcoder version
loopcoder init --repo .
loopcoder skill install --repo .
loopcoder doctor --repo .
loopcoder report --repo .
Then drive work from a conductor session:
/loopcoder <your need>
New init scaffolds default to adapters.gate: human-merge; pass
loopcoder init --repo . --gate auto only when the project should opt into
automatic production promotion.
Command side effects
| Command | Side effects |
|---|---|
loopcoder init --repo . |
Writes .delivery.yml, ROADMAP.md, GitHub labels, and local .git/info/exclude protection for .loopcoder/. |
loopcoder skill install --repo . |
Writes or refreshes global skill files, project hook settings, .loopcoder/conductor-workspace, and local .git/info/exclude protection. |
loopcoder doctor --repo . |
Read-only diagnostics; --format json emits machine-readable checks. |
loopcoder report --repo . |
Read-only local report query. |
loopcoder status --repo . |
Read-only local run status. |
loopcoder state push --repo . |
Explicitly writes run summaries to the dedicated state branch. |
loopcoder promote --repo . |
May change the configured production branch, subject to adapters.gate and the human command. |
What changed
loopcoder init --repo <path>and--gate human-merge|auto; new scaffolds
default tohuman-merge.initandskill install --repoprotect repo-local.loopcoder/through
.git/info/excludewithout touching tracked.gitignore.loopcoder doctor --format text|jsonincludes local-state, reportquery,
installed-skill, hook, provider, and migration checks.loopcoder report --format jsonnow includes both top-levelreportsand
richerrecordsentries withreport,source,run_id, andpath.- README and usage docs list every registered command, including
report,
state,lease,ps, andkill. - Pretty report docs now match the renderer's combined provider line, such as
OpenAI Codex / codex.
Local state boundary
.loopcoder/ is repo-local machine state for runs, relay ledgers, recovery,
status, and report queries. It must not be committed to normal business
branches. loopcoder state push is the explicit state-branch publishing path.
A machine can serve many projects: each project owns its own .delivery.yml
and .loopcoder/, while the machine-level binary and bundled skill live under
the user's machine-level loopcoder/agent directories.
v0.6.1 introduces no SQLite database, global project registry, project alias
map, or native sub-agent scheduler.
Compatibility
The 0.5.x to 0.6.x reporter compatibility window remains intact. Legacy inputs
and aliases are still accepted, while new output and new documentation use the
reporter terminology.