Improve agentic workflow — CI rules, branch strategy, deprecation rules, and new agents #28
Reference in New Issue
Block a user
Delete Branch "improve-agentic-workflow"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Summary
Improves the agentic workflow configuration with consistent rules across all agents.
Changes
issue-creator(mode: subagent) — orchestrator that dispatches sub-agents and creates GitHub issuesfeature-scout— scans for feature opportunities (TODOs, hardcoded values, rsync gaps)security-screener— scans for vulnerabilities (buffer overflows, TLS, path traversal)code-quality-guardian— scans for quality issues (god functions, duplication, complexity)Review
CI is green, new agents are well-structured, integrator fix correct. One stale section needs attention.
🔴 cmake-expert.md — stale "When Adding Sanitizer Support" section contradicts correct pattern
The document has two contradictory sanitizer instructions:
Lines 148–163 (correct, "Sanitizer Configurations"):
Lines 165–191 (wrong, "When Adding Sanitizer Support to CMakeLists.txt"):
This is the old stale pattern that was already flagged in the PR #26 review. The real CMakeLists.txt uses a single
SANITIZERcache variable withelseifbranches — not three separateoption()booleans. An agent that reads the "When Adding Sanitizer Support" section will produce wrong CMake code.Fix: Replace lines 165–191 with instructions showing how to add a new value to the existing
SANITIZERcache variable (e.g., add anelseif(SANITIZER STREQUAL "undefined")block). Or remove the section entirely — the correct pattern is already shown above and in AGENTS.md.🟡 Dockerfile — lcov/valgrind addition conflicts with PR #27's AGENTS.md rule
PR #27's AGENTS.md says "never add
apt-get installto CI workflows — use the custom Docker image". PR #25 adds lcov/valgrind to the Dockerfile. This PR also adds them. Both PRs are correct individually, but whoever merges second should confirm the Dockerfile is clean.No action needed from this PR — just flagging the overlap. The lcov/valgrind addition in the Dockerfile is the right approach per AGENTS.md.
New agents — well-structured
gh issue create. Usesmode: subagentcorrectly.All three scanner agents have the correct "No findings" return format, severity guidelines, and the standard three footer sections (CI, Branch Strategy, Dependency Installation).
integrator.md — fix correct
The CI example now uses
container: gitea.tap-tap.win/taptap/fastsync-ci:v7instead ofsudo apt-get install. Correct per AGENTS.md dependency rules.Verdict: Approve with 1 fix needed. The stale
ENABLE_ASAN/ENABLE_TSAN/ENABLE_UBSANsection in cmake-expert.md should be updated or removed — it contradicts the correct pattern shown earlier in the same file.Re-review: Approve ✅ (2 minor items)
The critical cmake-expert.md fix is clean — the stale
ENABLE_ASAN/ENABLE_TSAN/ENABLE_UBSANsection is gone, replaced with correctSANITIZERcache variable documentation. All other changes since last review are the PR #26 merge (ci.yaml, client_cli.c, reviewer.md) which we already approved.Verified fixes
ENABLE_ASANpatternSANITIZERcache variable +elseifguidanceRemaining minor items (non-blocking)
1. integrator.md line 110: inaccurate CI job description
The CI matrix only has
[address]. UBSan is not in the matrix. Should read "ASan build and test" or "Sanitizer build and test (address)".2. cmake-expert.md: UBSan workaround section
The manual
-DCMAKE_C_FLAGS="-fsanitize=undefined"workaround is documented with the accurate caveat "no-DSANITIZER=undefinedoption in CMakeLists.txt yet". This will become stale when PR #25 lands (which adds UBSan to the SANITIZER option + CI matrix), but is correct for now.Verdict: Approve. All previous critical items fixed. The two minor items above are non-blocking — the integrator.md inaccuracy is cosmetic (agents will still produce correct CI YAML from the example), and the UBSan workaround is correctly caveated.