feat(ci): use $RUNNER_TEMP instead of /tmp in claude-fix workflow #44

Merged

nrupard merged 2 commits from feat/pms-132-no-tmp into main

2026-05-27 17:20:45 +02:00

nrupard commented

2026-05-27 17:16:17 +02:00

Owner

Closes PMS-132.

The claude-fix workflow wrote three scratch files to hardcoded /tmp paths: claude-context/CONTEXT.md, claude-result.json, and commit-message.txt. This switches all of them to the job-scoped ${RUNNER_TEMP} directory that the Forgejo runner provides. RUNNER_TEMP is constant across steps within a job, so the context file written in the "Build failure context" step is still readable in the "Run Claude Code" step, and the runner cleans it up per job instead of leaving artifacts in a shared /tmp.

The "Run Claude Code" prompt heredoc was single-quoted and therefore could not expand a shell variable. It now resolves CONTEXT_FILE before an unquoted heredoc; the prompt body contains no other shell metacharacters, so unquoting is safe.

No other workflow or runtime code uses hardcoded /tmp. The remaining temp usage in the tree (tempfile crate in tests / src/infisical/dev.rs, mktemp --tmpdir in the justfile) honors $TMPDIR and uses random unique names, so it is out of scope for this change.

Closes PMS-132. The `claude-fix` workflow wrote three scratch files to hardcoded `/tmp` paths: `claude-context/CONTEXT.md`, `claude-result.json`, and `commit-message.txt`. This switches all of them to the job-scoped `${RUNNER_TEMP}` directory that the Forgejo runner provides. `RUNNER_TEMP` is constant across steps within a job, so the context file written in the "Build failure context" step is still readable in the "Run Claude Code" step, and the runner cleans it up per job instead of leaving artifacts in a shared `/tmp`. The "Run Claude Code" prompt heredoc was single-quoted and therefore could not expand a shell variable. It now resolves `CONTEXT_FILE` before an unquoted heredoc; the prompt body contains no other shell metacharacters, so unquoting is safe. No other workflow or runtime code uses hardcoded `/tmp`. The remaining temp usage in the tree (`tempfile` crate in tests / `src/infisical/dev.rs`, `mktemp --tmpdir` in the justfile) honors `$TMPDIR` and uses random unique names, so it is out of scope for this change.

nrupard added 1 commit

2026-05-27 17:16:17 +02:00

feat(ci): use $RUNNER_TEMP instead of /tmp in claude-fix workflow

Check / fmt + clippy + compile + tests (pull_request) Has been cancelled

Details

54f194189a

PMS-132. The claude-fix workflow wrote scratch files (claude-context/CONTEXT.md, claude-result.json, commit-message.txt) to hardcoded /tmp paths. Switch all of them to the job-scoped ${RUNNER_TEMP} directory that the Forgejo runner provides. RUNNER_TEMP is constant across steps within a job, so the context file written in one step is still readable in later steps, and it is cleaned up automatically per job rather than persisting in a shared /tmp.

The Run Claude Code step's prompt heredoc was single-quoted, so it could not expand a shell variable; it now resolves CONTEXT_FILE before an unquoted heredoc (the prompt body contains no other shell metacharacters).

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

nrupard added 1 commit

2026-05-27 17:18:25 +02:00

feat(ci): guard against unset RUNNER_TEMP before using it

Check / fmt + clippy + compile + tests (pull_request) Successful in 28s

Details

Create release / Create release from merged PR (pull_request) Has been skipped

Details

bcd0990be9

Add `: "${RUNNER_TEMP:?...}"` at the top of the three run blocks that dereference RUNNER_TEMP. If the runner ever fails to set it, the step now fails fast with a clear message instead of silently writing scratch files to the container filesystem root.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>