Resolve #74 merge conflict: sync .agent issue-fns.sh mirror with master (jira/ado)

[laird]

Unblocks #74 (issue #71), which is currently CONFLICTING and cannot merge.

Base is feature/issue-71, not master — merging this into that branch clears the conflict without force-pushing over #74's history.

The conflict was semantic, not mechanical

One file conflicted: .agent/scripts/issue-fns.sh.

• Fix #71: resync the .agent mirror issue layer; add a parity guard #74's side replaces the stale fat mirror with the thin 9-verb dispatcher. That is the entire point of .agent mirror issue layer is stale: workflows call issue_claim/issue_release/issue_any_claimable which do not exist #71 — .agent/workflows/fix.md called issue_claim / issue_release / issue_any_claimable, which the old mirror never defined, so Antigravity's /fix exited 127 at preflight on every run.
• master's side still has the fat implementation, but has since gained jira and ado backends (_ifns_jira, _ifns_ado) from Add Jira and Azure DevOps issue backends (#81, #82) #83.

Resolving "ours" would have silently deleted Jira and Azure DevOps support from the .agent mirror. Worth flagging, because both sides look correct in isolation and the loss would not show up as a test failure on #74 alone.

Resolution

Took the canonical plugins/autocoder/scripts/issue-fns.sh from master verbatim — it is already the thin dispatcher and carries the jira / ado cases. .agent/scripts/ ships issues-jira.sh and issues-ado.sh, so those cases resolve.

This is not a judgment call: tests/test_agent_mirror_parity.sh (added by #74 itself) requires the shared files to be byte-identical across mirrors, which makes the canonical file the only valid resolution — and self-verifying.

Verification

$ bash tests/test_agent_mirror_parity.sh
Results: 13 passed, 0 failed      # incl. "issue-fns.sh identical across mirrors"
                                  # and "mirror issue-fns.sh defines all 9 contract verbs"

$ bash -n plugins/autocoder/scripts/*.sh && bash -n .agent/scripts/*.sh \
    && python3 -m py_compile plugins/autocoder/scripts/*.py
✅ syntax OK

$ for t in tests/test_*.sh; do bash "$t"; done
18/18 suites pass, 0 failures

Suites verified directly rather than through regression-test.sh: this branch predates #73, so the runner here can still report a false green and is not trustworthy as the gate.

Merge order

1. this PR → feature/issue-71
2. Fix #71: resync the .agent mirror issue layer; add a parity guard #74 → master (now conflict-free)

🤖 Generated with Claude Code

[laird]

Both dry-run findings folded in. Branch updated: b214e16.

1. #87 did not fully clear #74 — fixed

Confirmed, and the cause was mine. api-push.py builds commits with parents=[base_sha] — a single parent — so it flattened my real two-parent merge into one commit. Master was therefore not in the branch's ancestry, and git re-derived the identical conflict on the later merge to master:

$ git merge-tree $(git merge-base origin/merge/issue-71-master origin/master) \
    origin/merge/issue-71-master origin/master
changed in both
  base 6530e5e .agent/scripts/issue-fns.sh
  our  bd1194d
  their e8f38a3

The file content was already correct; only the ancestry was wrong. Rebuilt the branch through the Git Data API so the merge is recorded with both parents (c714eb9 = feature/issue-71, a4cc38b = master) rather than flattened.

$ git merge-base --is-ancestor origin/master origin/merge/issue-71-master && echo YES
YES

2. Mirror was short issue-source-lib.sh — fixed

Confirmed absent. It is issue-source resolution used by the swarm launch/lifecycle scripts — shared issue-layer infrastructure — so it belongs in the mirror. Copied verbatim.

Also widened the guard's SHARED list from 4 files to the whole issue layer (7): issues-jira.sh and issues-ado.sh arrived with #83 and were never checked, and neither was issue-source-lib.sh. A backend that ships only under plugins/ is invisible to this guard — which is precisely how the drift in #71 went unnoticed. The mirror looked healthy because nothing checked the files missing from it.

Acceptance

# 1. full chain merges clean
$ git merge origin/merge/issue-71-master     # -> feature/issue-71 : clean
$ git merge origin/master                    # -> Already up to date.
conflicts remaining: 0

# 2. parity guard covers the whole layer, verified in the MERGED state
$ bash tests/test_agent_mirror_parity.sh
Results: 19 passed, 0 failed        # was 13/13

$ for t in tests/test_*.sh; do bash "$t"; done
SUITES FAILED: 0 / 18

$ bash -n plugins/autocoder/scripts/*.sh && bash -n .agent/scripts/*.sh \
    && python3 -m py_compile plugins/autocoder/scripts/*.py
✅ syntax OK

Verification ran in a scratch worktree at the merged result, not on my branch, so it reflects what actually lands.

The new assertions were negative-tested rather than assumed: removing issue-source-lib.sh from the mirror fails 2 assertions, and perturbing issues-jira.sh fails the divergence check. They bite.

Still noted, not fixed here

.agent/scripts/start-parallel-agents.sh does not source issue-source-lib.sh, while master's plugins/ copy does — so that script's mirror is itself behind. Same drift class, but out of scope for this PR and it does not affect either acceptance criterion. Worth its own issue.

Merge order unchanged: this PR → feature/issue-71, then #74 → master.

🤖 Autonomous fix workflow