A refused handoff had no feedback loop. PrepareRelease read the report, the
parser refused it, the worker recorded the error in health, and the next
boundary read the same bytes and refused them again. Run 10 spent four leases
that way and the agent was never told anything.
The plan-progress path already had the answer: answerRefusedProgress says why,
drops the file, and lets the agent write a corrected one. The release path now
does the same, gated on a typed ErrInvalidHandoffAnswer so a transport or Git
failure keeps its retry.
This is the silent-loop shape CLAUDE.md names, in a path nobody had checked.
The three format fixes above it each removed one trigger; this removes the
loop.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CVbaKucEYBjMqVeUgJUsc1
The prompt gave "at most 200 characters" for NEXT, WHY and REMAINING. The
validator applies it to OPEN Q and LEARNED as well, and run 10 lost a lease to
a 219-character OPEN Q against a limit nobody had stated for that field.
Third instance of one shape: a constraint the code enforces and no brief
mentions. The other two were the research finding id and the dead-end
separator.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CVbaKucEYBjMqVeUgJUsc1
98f1b2d fixed the arrow and left the other half. "tried" and "failed because"
restate the field names either side of a separator that already says what the
line means, and the parser required both.
Run 9 lost three leases to this. Every refused line carried exactly the data
the fields want:
tried run line ["bash", "scripts/test_healthcheck.sh"] for phase-3
-> refused, policy allows only bash -n on one file
tried finding ids "F1".."F8" in research.json
-> schema requires lowercase letters, digits, dash or underscore
Both prefixes are now optional. A line with no cause and effect is still
refused, because the split is the thing being validated.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CVbaKucEYBjMqVeUgJUsc1
Run 9's planner wrote a dead end as "tried X -> failed because Y". The content
was exactly what the protocol asks for. The parser split on "→" only, refused
the handoff, and a refused handoff fails the release rather than the turn, so
the lease expired with the work intact and unrecorded.
Accept either arrow. A separator is not the thing being validated.
The prefix stripping was also wrong on the right-hand side: it trimmed
"failed because " from an untrimmed string that started with a space, so every
parsed dead end kept the literal prefix in why_failed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CVbaKucEYBjMqVeUgJUsc1
Handoff.Command is the last command observed in the pane. The handoff prompt
tells the agent to write .orchestra-handoff-report.md and stop, so that write
is almost always the last command there. Validate then rejects it as circular:
"must not point to a handoff or report".
Every phase rotation therefore failed on Orchestra's own instruction. Live on
run 5, one stage past F36.
lastObservedCommand now skips commands the validator would call circular, so
the successor gets the last command that was real work. The rule stays in
continuity and is asked, not restated, which is how F36 happened.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011xsXyr5J1RACo71YeKG3Pu
handoffReason has emitted "phase_changed" since phase rotations landed. The
continuity validator's reason list was never extended, so every phase rotation
built a handoff it then refused as "invalid handoff meta".
Live on run 5: with F31 clearing the parse ahead of it, the release reached
this and stopped at phase "prepared" with
"adapter: upload handoff: invalid handoff meta".
The test asserts the property rather than the constant: every reason the
adapter can produce must survive Validate, including its fallback.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011xsXyr5J1RACo71YeKG3Pu
Release now re-verifies every handoff Anchor.Dirty file hash (previously
unchecked after the top-level anchor SHA compare), snapshots dirty state
onto a per-task scratch branch before uploading, and rewrites the anchor
to the new commit so successor pickup collapses to a single HEAD compare.
ScratchCommit made idempotent for repeated rotations of the same task.
Bootstrap's prompt now points the agent at the scratch-branch commit
history instead of vague "read the handoff" prose, and does not claim a
GET /v1/artifacts/<ref> endpoint that doesn't exist.
MarkdownChanges had zero callers and zero tests; deleted per AUDIT.md's
explicit deletion option rather than half-wiring an undesigned feature.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01W1rkJ2hBMybnJctPbcy4tT
kami
kami