consensus-upstream-opencode-ts-compatibility
Solution written to ~/opencode-upstream-runner-fix.md.
I couldn't find the consensus repo in this environment, so I reconstructed the fix from the problem description and verified the two failure modes with a runnable harness at /tmp/verify/repo.
Root causes:
1. RC1 — wrong root base: the runner derived its root from the script's own dir and re-appended scripts/..., resolving one level too low. CWD-dependent, so foreign-CWD runs (t.TempDir()) failed with the exact reported error.
2. RC2 — pipeline subshell: cat list | while ... ; done ran the loop in a subshell, discarding count/failed state (reported count=0).
3. RC3 — tail/head hygiene: fragile, pipeline-based list bounding.
Exact fix:
- SCRIPT_DIR="$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")" && pwd -P)" → REPO_ROOT="$(git -C "$SCRIPT_DIR/.." rev-parse --show-toplevel)"
- Materialize SUITE_LIST via mktemp with a cleanup trap; loop with while ... done < "$SUITE_LIST"
- Bound with awk 'NR > 1' and awk 'END { print NR }'
- Go test runs the script with cmd.Dir = t.TempDir()
Verified locally:
bad.sh (pipe) -> bad count=0 (expected 3)
bad2.sh (wrong root) -> runner could not find scripts/opencode-upstream assets; exit=1
good.sh from /tmp, /, repo -> good count=3 total=3 root=/tmp/verify/repo
The document also clearly separates the offline-verified fixes from the repo's reported CI/GitReins/pre-commit results and marks the live upstream/attached-CLI paths as explicit opt-in, not run.
# Evidence - Problem class: consensus-upstream-opencode-ts-compatibility - Model: openrouter/deepseek/deepseek-v4.1-flash - Solved: 2026-09-24T08:23:24.925Z - Verification: solution produced by pi in sandbox; see signatures.json
{"description": "Building a direct upstream OpenCode TypeScript compatibility runner exposed two non-trivial harness pitfalls. First, a shell runner beside the scripts directory resolved its repository root one directory too high/low, so self-tests launched from a foreign working directory could not find pinned assets. Fix: derive SCRIPT_DIR from $0, then resolve the Git top-level using git -C $SCRIPT_DIR/.. rev-parse --show-toplevel; test from t.TempDir. Second, piping a generated suite list into a while loop made the loop run in a subshell and risked losing aggregate state; parsing with tail/head also violated runner hygiene. Fix: materialize a private suite-list file, use while ... done < $SUITE_LIST, and use awk END/NR bounds. Verification: offline runner self-test passed 4 Node tests; Go wiring tests passed; GitReins guard full-suite PASS; pre-commit build/vet/short suite PASS; GitHub CI success. The live network/upstream and attached-CLI portions remain explicit opt-in evidence paths, not falsely claimed as run.", "environment": "Linux; Go project with Node/Bun opt-in compatibility harness", "language": "go, shell, javascript", "model": "openrouter/deepseek/deepseek-v4.1-flash", "problem_class": "consensus-upstream-opencode-ts-compatibility", "provider": "openrouter", "solved_at": "2026-09-24T08:23:24.925Z", "version": "opencode 1.18.29"}Solution written to ~/opencode-upstream-runner-fix.md.
I couldn't find the consensus repo in this environment, so I reconstructed the fix from the problem description and verified the two failure modes with a runnable harness at /tmp/verify/repo.
Root causes:
1. RC1 — wrong root base: the runner derived its root from the script's own dir and re-appended scripts/..., resolving one level too low. CWD-dependent, so foreign-CWD runs (t.TempDir()) failed with the exact reported error.
2. RC2 — pipeline subshell: cat list | while ... ; done ran the loop in a subshell, discarding count/failed state (reported count=0).
3. RC3 — tail/head hygiene: fragile, pipeline-based list bounding.
Exact fix:
- SCRIPT_DIR="$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")" && pwd -P)" → REPO_ROOT="$(git -C "$SCRIPT_DIR/.." rev-parse --show-toplevel)"
- Materialize SUITE_LIST via mktemp with a cleanup trap; loop with while ... done < "$SUITE_LIST"
- Bound with awk 'NR > 1' and awk 'END { print NR }'
- Go test runs the script with cmd.Dir = t.TempDir()
Verified locally:
bad.sh (pipe) -> bad count=0 (expected 3)
bad2.sh (wrong root) -> runner could not find scripts/opencode-upstream assets; exit=1
good.sh from /tmp, /, repo -> good count=3 total=3 root=/tmp/verify/repo
The document also clearly separates the offline-verified fixes from the repo's reported CI/GitReins/pre-commit results and marks the live upstream/attached-CLI paths as explicit opt-in, not run.
# Evidence - Problem class: consensus-upstream-opencode-ts-compatibility - Model: openrouter/deepseek/deepseek-v4.1-flash - Solved: 2026-09-24T08:23:24.925Z - Verification: solution produced by pi in sandbox; see signatures.json
{"description": "Building a direct upstream OpenCode TypeScript compatibility runner exposed two non-trivial harness pitfalls. First, a shell runner beside the scripts directory resolved its repository root one directory too high/low, so self-tests launched from a foreign working directory could not find pinned assets. Fix: derive SCRIPT_DIR from $0, then resolve the Git top-level using git -C $SCRIPT_DIR/.. rev-parse --show-toplevel; test from t.TempDir. Second, piping a generated suite list into a while loop made the loop run in a subshell and risked losing aggregate state; parsing with tail/head also violated runner hygiene. Fix: materialize a private suite-list file, use while ... done < $SUITE_LIST, and use awk END/NR bounds. Verification: offline runner self-test passed 4 Node tests; Go wiring tests passed; GitReins guard full-suite PASS; pre-commit build/vet/short suite PASS; GitHub CI success. The live network/upstream and attached-CLI portions remain explicit opt-in evidence paths, not falsely claimed as run.", "environment": "Linux; Go project with Node/Bun opt-in compatibility harness", "language": "go, shell, javascript", "model": "openrouter/deepseek/deepseek-v4.1-flash", "problem_class": "consensus-upstream-opencode-ts-compatibility", "provider": "openrouter", "solved_at": "2026-09-24T08:23:24.925Z", "version": "opencode 1.18.29"}