amirfish1 / pair-verify

Pair-verify spawns a single skeptic CCC session whose job is to disprove
your fix. It reproduces the original bug on a clean worktree at the pre-fix
commit, confirms the bug is real, then checks out the fix ref and proves the
behavior actually changed — plus one short pass hunting for a regression the
fix could introduce. It builds on the spawn/report_to mechanics in the
ccc-orchestration skill — read that first for the full Spawn/Inject/Ask API.

Cost

1 spawned session. Each spawn is a real billed session on the user's
kanban — tell the user the count before spawning. Do not use this for
trivial fixes that already ship with a test proving the fix (just run the test —
a green test on the fix and red on the pre-fix ref is the same evidence, for
free).

Setup

CCC_URL="$(cat ~/.claude/command-center/port.txt 2>/dev/null || echo "${CCC_URL:-http://127.0.0.1:8090}")"
curl -sf --max-time 2 "$CCC_URL/api/version" >/dev/null 2>&1 || CCC_URL="http://127.0.0.1:8090"
curl -sf --max-time 2 "$CCC_URL/api/version" >/dev/null 2>&1 || CCC_DOWN=1

• (a) Run every CCC curl with the network sandbox disabled — the Bash
  sandbox blocks loopback and fails spuriously even when CCC is up.

• (b) URL-encode repo_path in query strings (+ → %2B, space → %20).

• (c) Resolve your own session id: $CLAUDE_SESSION_ID, else the newest
  *.jsonl under ~/.claude/projects/<slugified-cwd>/ (see ccc-orchestration).

Preconditions — gather these BEFORE spawning

The skeptic can only reproduce what you can describe. Have all three ready:

• (a) Exact repro steps for the original bug: the commands / inputs, and
  expected-vs-actual output. "It was broken" is not repro steps.

• (b) The pre-fix ref and the fix ref — commit shas (not "before" / "after").
• (c) The repo path.

If you cannot state the repro steps concretely, you are not ready to verify —
go get them first. A verifier with no reproducible bug verifies nothing.

Spawning the skeptic

Resolve your session id, fill in the shas and steps, then POST one spawn:

curl -s -X POST "$CCC_URL/api/sessions/spawn" -H "Content-Type: application/json" -d @- <<'JSON'
{
  "prompt": "Your job is to prove this fix does NOT work. Be adversarial. 1) Create a clean worktree at the pre-fix ref: git worktree add /tmp/pairverify-<slug> <pre-fix-ref>. 2) Run the repro steps: <exact commands / inputs, with expected-vs-actual>. Confirm the ORIGINAL bug reproduces. If it does NOT reproduce, STOP and report VERDICT: REPRO-FAILED — that alone is a major finding. 3) In the worktree, check out <fix-ref>, run the SAME repro, and confirm it now behaves correctly. 4) Spend one short pass hunting a plausible regression the fix could introduce (adjacent callers, edge inputs, boundary cases). 5) Clean up: git worktree remove /tmp/pairverify-<slug>. Report back: VERDICT (VERIFIED / REPRO-FAILED / FIX-INCOMPLETE / REGRESSION-RISK), plus the exact commands you ran and their output as evidence.",
  "repo_path": "/abs/path/to/repo",
  "report_to": "<your-session-id>"
}
JSON

Omit "model" and "reasoning_effort": the server spawn defaults (Settings →
Spawn defaults) apply. Pass either explicitly to keep cost down, with
"engine" alongside it, since both are validated per engine (see
ccc-orchestration § 2 for the per-engine effort ladders). CCC appends the
return-address footer (report_to), so the skeptic injects its verdict back to
you when it finishes.

Waiting for the report

After spawning, end your turn. Tell the user a skeptic is verifying the fix
and give its session id. The verdict arrives by injection (report_to) — do
not poll or sleep-loop. If it never arrives, check
GET /api/sessions/spawned for whether the child is still alive.

Interpreting the verdict

• VERIFIED — the bug reproduced at the pre-fix ref and is gone at the fix
  ref. Ship it. If it came with a regression note, that still needs your
  judgment before you claim done.

• REPRO-FAILED — the skeptic could not reproduce the original bug. This
  means your understanding of the bug is likely wrong. Do not ship on it —
  figure out what the real bug was.

• FIX-INCOMPLETE — the bug still shows at the fix ref. Back to debugging.

• REGRESSION-RISK — the fix works but the skeptic found a plausible new
  break. Investigate before shipping.

Dry run

If the arguments contain dry-run, print the exact payload — session count (1),
the full skeptic prompt with the shas and repro steps filled in, and the target
repo — and POST nothing.

When CCC is down

If CCC_DOWN=1 or the spawn curl returns connection refused (exit 7), run the
same reproduce-first protocol yourself in a fresh git worktree. In your final
claim, say verification was self-run — you know too much about the fix to be
a real skeptic (same-context bias), so flag it. Never pretend the CCC spawn ran.

What this is not

• Not the in-process /verify pass (that's you driving your own change). This
  buys an independent skeptic with reproduce-first discipline.

• Not /code-review — that reads the diff. This actually runs the bug at two
  refs and compares behavior.