The harness declared peer deps on the pre-rename @mariozechner/pi-ai and
@mariozechner/pi-agent-core names, so npm auto-installed our own code from the
registry under an alias. sync-versions bumped those aliases in lockstep, which
made every release fail because the new version is not published yet.
Build the eval harness on createHarness from vitest-evals core instead. The
pi-specific code was already ours; the harness only contributed agent-lifecycle
plumbing and unused tool replay.
Add missing entries for the llama.cpp catalog cache and resource loader
directory fix, duplicate ai-side changes into coding-agent, add missing
external PR attributions, restore the 0.82.0 section that was modified
after release, and add the New Features section.
Only comment the analysis summary when the pi run succeeded, add an explicit
failure comment, skip unchanged auth secret writes, and document that
PI_AUTH_UPDATE_TOKEN needs the Environments permission.
The /copy command always set copied=true after spawning wl-copy without
awaiting its exit code. When wl-copy fails (e.g. in a sandboxed
environment that inherits WAYLAND_DISPLAY but not the socket), the
xclip/OSC 52 fallbacks never ran because copied was already true.
Fix: await the wl-copy spawn exit code and only claim success on exit
code 0. On failure, fall through to the x11 clipboard fallback.
Fixes#6872.
Add the constrained sampling API and provider plumbing for OpenAI and Anthropic tool calls, including grammar sampling variants and model capability metadata.
Maintainers frequently direct lgtm at tagged users who volunteer on an issue, but approval previously applied to the issue or PR author. Mentioned users are now approval targets, with the author as the fallback.
Prose and examples can also contain lgtm without intending approval. Only treat lgtm or lgtmi at the start, after leading mentions, or at the end as commands. Process PR comments too, where approvals previously had no effect.
Fresh package installs could give bundled model catalog
files newer filesystem mtimes than the remote catalog’s Last-Modified timestamp,
causing Pi to ignore newer remote models.
Use the catalog’s recorded generation time instead of installation-dependent file metadata.
Replace OpenAI and Anthropic SDK retries with local request retries that
preserve configured retry counts, enforce maxRetryDelayMs, and make retry
backoff interruptible through AbortSignal.
Alternatives considered: Racing SDK requests against AbortSignal would make
cancellation observable but would not let Pi enforce maxRetryDelayMs before
the SDK enters its retry sleep. Patching private SDK internals would be brittle,
while disabling provider retries entirely would ignore explicitly configured
maxRetries.
Also, align the custom OpenAI Codex retry loop with the documented
maxRetryDelayMs behavior. Oversized server-requested delays now fail immediately
instead of being clamped, allowing the visible, abortable agent-level retry
policy to handle the failure.
Fixes#6911
