You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
factory watch observed a runtime failure of the factory itself from local evidence: signal lifecycle, stage manage, pr_frontier_failed. Filed automatically; no human has looked at it yet.
Triage: This is a fully specified bug report: it has a clear problem statement (factory runtime failure at the manage stage with mechanism failure pr_frontier_failed, fingerprint b4b6c83c4478), concrete reproduction steps, and an observable done-condition (factory pass completes without pr_frontier_failed and factory watch --dry-run no longer reports the fingerprint). It is contained within this repository and does not touch release, signing, or security policy, so it does not require human authorization or design judgment. An agent can investigate the manage-stage mechanism failure, fix the root cause, and verify via the provided commands.
Agent brief: The factory's manage stage exits with outcome mechanism_failure and reason pr_frontier_failed (fingerprint b4b6c83c4478), detected automatically via factory watch. Investigate the root cause of the manage-stage mechanism failure and fix it so the factory pass completes cleanly without that failure. Verification: zcat -f .factory/events.jsonl* | grep -F '"reason":"pr_frontier_failed"' returns nothing, and factory watch --dry-run no longer reports fingerprint b4b6c83c4478.
Not a rocm-cli bug. One manage execution on 2026-10-02T13:25:57Z for ticket 16
raised inside factory/manage.py::frontier_pass: its first gh child
(gh issue view 16 --json …, pid 2612588) exited non-zero after 0.39 s and the
handler at manage.py:746-749 set mechanism_failure / pr_frontier_failed.
The next pass completed; the retained journal has 7009 completed manage exits
for ticket 16 and this single failure, so the gh error was transient. Its
text was only ever printed to the dispatcher log (str(CalledProcessError)
carries no stderr), and the journal for that boot has rotated out, so the
specific GitHub error is unrecoverable.
The factory defect that made this a distinct ticket: the frontier was the only
handler minting its own reason. pr_frontier_failed is absent from factory/runtime_events.py::REASONS (so dashboard --runtime-json sanitises
it to null), while the viability, escalation, handoff and triage handlers all
journal the same failure class as github_command_failed — the fingerprint factory watch already tracks for it.
What changed
Fix lives in the factory repo, not this one — nothing to commit on agent/52.
Repo: ~/.local/share/district/roots/factory, worktree .factory/wt-rocm-cli-52, branch agent/rocm-cli-52, one signed-off commit
on origin/main (51f9e03, == released factory 0.4.1).
tests/test_frontier.py::test_failed_gh_read_journals_the_shared_github_command_failed_reason:
a failing gh issue view in the frontier journals ("mechanism_failure", "github_command_failed") and that reason is in runtime_events.REASONS.
Failed before (observed pr_frontier_failed), passes after.
Against the real rocm-cli journal with the patched code on PYTHONPATH: factory watch --dry-run does not report b4b6c83c4478 (it also does not
with the installed tool, because factory runtime failure: lifecycle manage pr_frontier_failed #52 is the filing for it and no newer
evidence exists).
rocm-cli gate: see .factory/gate-report-52.md.
Full factory suite (unittest discover -s tests -t .): 588 run, 2 errors — tests.test_mcp and tests.test_plan fail to import (No module named 'test_evidence'); identical on a detached origin/main worktree, so
pre-existing and unrelated.
Unverified / not done
The installed uv tool factory was not reinstalled; the running dispatcher
keeps the old reason until the factory branch merges and the tool upgrades.
zcat -f .factory/events.jsonl* | grep -F '"reason":"pr_frontier_failed"'
still returns the one 2026-10-02 row: that is the journal, not code, and was
left intact. After the upgrade no new row with that reason can be written.
All four gh-failure handlers log str(exc) without exc.stderr, so the next
transient gh failure will again be undiagnosable once the journal rotates.
Out of scope here; one-line improvement if it recurs.
Next
Push agent/rocm-cli-52 in the factory repo and open a PR there.
After merge: upgrade the installed tool, then factory watch --dry-run in
rocm-cli to confirm with the installed code.
Close this ticket as fixed-upstream; no rocm-cli change.
Factory manager: The escalation is a repository mismatch, not a failed rocm-cli gate: the worker changed the factory repository and left no commits on agent/52, so the dispatcher had nothing to publish here.
The external worktree’s factory/manage.py:748 uses github_command_failed. This normalizes the journal reason; it does not prevent GitHub command failures. The worker reports a transient failure, but its underlying stderr is unavailable. Absence of the fingerprint from watch --dry-run is insufficient proof: the handoff says the unpatched tool also omits it because #52 already tracks the evidence.
Human action: review and publish agent/rocm-cli-52 from the factory repository, verify its checks, then merge and upgrade the installed factory. Confirm the installed behavior before closing #52. Publication, merge, and installation status were not verified here. Retrying the rocm-cli worker would target the wrong repository again.
Factory handoff request 52/1: a human decision is needed on #52.
No decision owner could be determined (unassigned); a maintainer should claim it.
Question
How should #52 proceed? It escalated with: agent/52: PR not published (no commits over main or existing PR target mismatch); inspect dispatcher log. Automatic recovery stopped because the manager asked for a human decision.
Evidence (bounded links; worker logs and the escalation packet stay on the runner)
Proposed next step
Read the evidence, then amend the ticket and relabel ready-for-agent for another attempt, fix the branch by hand, request changes on the PR, or close it. Replying here is context for humans; nothing here retries, approves or merges.
Routing
decision_owner: absent — ticket body has no **Decision owner** section
initiative: skipped — initiative owner only routes requirements, not implementation
configuration: absent — no [collaboration] section in .factory.toml; legacy behaviour unchanged
To claim or reassign this decision, set a **Decision owner** section (a GitHub login) in the issue body; the next pass reflects it and never overwrites it.
Duplicate of mikeroySoft/factory#264 (manager gh failures; host gh failures are no longer filed since ROCm#281). Stopping watch from filing engine bugs here is tracked in mikeroySoft/factory#311. Engine bugs are tracked in mikeroySoft/factory.
Activity
mikeroysoft-agent commentedon Oct 17, 2026
Triage: This is a fully specified bug report: it has a clear problem statement (factory runtime failure at the manage stage with mechanism failure pr_frontier_failed, fingerprint b4b6c83c4478), concrete reproduction steps, and an observable done-condition (factory pass completes without pr_frontier_failed and factory watch --dry-run no longer reports the fingerprint). It is contained within this repository and does not touch release, signing, or security policy, so it does not require human authorization or design judgment. An agent can investigate the manage-stage mechanism failure, fix the root cause, and verify via the provided commands.
Agent brief: The factory's manage stage exits with outcome mechanism_failure and reason pr_frontier_failed (fingerprint b4b6c83c4478), detected automatically via factory watch. Investigate the root cause of the manage-stage mechanism failure and fix it so the factory pass completes cleanly without that failure. Verification:
zcat -f .factory/events.jsonl* | grep -F '"reason":"pr_frontier_failed"'returns nothing, andfactory watch --dry-runno longer reports fingerprint b4b6c83c4478.mikeroysoft-agent commentedon Oct 17, 2026
Factory dispatcher escalating: agent/52: PR not published (no commits over main or existing PR target mismatch); inspect dispatcher log.
Escalation packet:
/home/mike/.local/share/district/roots/rocm-cli/.factory/escalations/52.mdWorker logs:
/home/mike/.local/share/district/roots/rocm-cli/.factory/logs/52-attempt-1.logWorker handoff notes:
Handoff: #52 factory runtime failure: lifecycle manage pr_frontier_failed
Root cause
Not a rocm-cli bug. One
manageexecution on 2026-10-02T13:25:57Z for ticket 16raised inside
factory/manage.py::frontier_pass: its firstghchild(
gh issue view 16 --json …, pid 2612588) exited non-zero after 0.39 s and thehandler at
manage.py:746-749setmechanism_failure/pr_frontier_failed.The next pass completed; the retained journal has 7009 completed
manageexitsfor ticket 16 and this single failure, so the
gherror was transient. Itstext was only ever printed to the dispatcher log (
str(CalledProcessError)carries no stderr), and the journal for that boot has rotated out, so the
specific GitHub error is unrecoverable.
The factory defect that made this a distinct ticket: the frontier was the only
handler minting its own reason.
pr_frontier_failedis absent fromfactory/runtime_events.py::REASONS(sodashboard --runtime-jsonsanitisesit to
null), while the viability, escalation, handoff and triage handlers alljournal the same failure class as
github_command_failed— the fingerprintfactory watchalready tracks for it.What changed
Fix lives in the factory repo, not this one — nothing to commit on
agent/52.~/.local/share/district/roots/factory, worktree.factory/wt-rocm-cli-52, branchagent/rocm-cli-52, one signed-off commiton
origin/main(51f9e03, == releasedfactory 0.4.1).factory/manage.py:748: reasonpr_frontier_failed→github_command_failed.tests/test_frontier.py::test_failed_gh_read_journals_the_shared_github_command_failed_reason:a failing
gh issue viewin the frontier journals("mechanism_failure", "github_command_failed")and that reason is inruntime_events.REASONS.Failed before (observed
pr_frontier_failed), passes after.CHANGELOG.mdUnreleased line.Verified
PYTHONPATH=. …/factory/bin/python -P -m unittest tests.test_frontier tests.test_watch tests.test_handoff tests.test_lifecycle_consumers→ 52 OK.uvx ruff@0.16.10 check factory/manage.py tests/test_frontier.py→ clean.PYTHONPATH:factory watch --dry-rundoes not reportb4b6c83c4478(it also does notwith the installed tool, because factory runtime failure: lifecycle manage pr_frontier_failed #52 is the filing for it and no newer
evidence exists).
.factory/gate-report-52.md.unittest discover -s tests -t .): 588 run, 2 errors —tests.test_mcpandtests.test_planfail to import (No module named 'test_evidence'); identical on a detachedorigin/mainworktree, sopre-existing and unrelated.
Unverified / not done
uv toolfactory was not reinstalled; the running dispatcherkeeps the old reason until the factory branch merges and the tool upgrades.
zcat -f .factory/events.jsonl* | grep -F '"reason":"pr_frontier_failed"'still returns the one 2026-10-02 row: that is the journal, not code, and was
left intact. After the upgrade no new row with that reason can be written.
str(exc)withoutexc.stderr, so the nexttransient
ghfailure will again be undiagnosable once the journal rotates.Out of scope here; one-line improvement if it recurs.
Next
agent/rocm-cli-52in the factory repo and open a PR there.factory watch --dry-runinrocm-cli to confirm with the installed code.
mikeroysoft-agent commentedon Oct 17, 2026
Factory manager: The escalation is a repository mismatch, not a failed rocm-cli gate: the worker changed the factory repository and left no commits on
agent/52, so the dispatcher had nothing to publish here.The external worktree’s
factory/manage.py:748usesgithub_command_failed. This normalizes the journal reason; it does not prevent GitHub command failures. The worker reports a transient failure, but its underlying stderr is unavailable. Absence of the fingerprint fromwatch --dry-runis insufficient proof: the handoff says the unpatched tool also omits it because #52 already tracks the evidence.Human action: review and publish
agent/rocm-cli-52from the factory repository, verify its checks, then merge and upgrade the installed factory. Confirm the installed behavior before closing #52. Publication, merge, and installation status were not verified here. Retrying the rocm-cli worker would target the wrong repository again.mikeroysoft-agent commentedon Oct 17, 2026
Factory handoff request
52/1: a human decision is needed on #52.No decision owner could be determined (unassigned); a maintainer should claim it.
Question
How should #52 proceed? It escalated with: agent/52: PR not published (no commits over main or existing PR target mismatch); inspect dispatcher log. Automatic recovery stopped because the manager asked for a human decision.
Evidence (bounded links; worker logs and the escalation packet stay on the runner)
Proposed next step
Read the evidence, then amend the ticket and relabel
ready-for-agentfor another attempt, fix the branch by hand, request changes on the PR, or close it. Replying here is context for humans; nothing here retries, approves or merges.Routing
**Decision owner**sectionTo claim or reassign this decision, set a
**Decision owner**section (a GitHub login) in the issue body; the next pass reflects it and never overwrites it.mikeroySoft commentedon Oct 19, 2026
Duplicate of mikeroySoft/factory#264 (manager gh failures; host gh failures are no longer filed since ROCm#281). Stopping watch from filing engine bugs here is tracked in mikeroySoft/factory#311. Engine bugs are tracked in mikeroySoft/factory.