fix: hashlib FIPS crash + assert→runtime guards in 4 production files

[AlexFucuson9]

Summary

Fixes two categories of bugs found during codebase scan:

1. hashlib FIPS crash (P1)

Two files call hashlib.md5() / hashlib.sha1() without usedforsecurity=False. On FIPS-enabled systems (RHEL 9, Ubuntu 22.04 FIPS, AWS AL2023 FIPS), these calls raise ValueError: [digital envelope routines: EVP_DigestInit_ex] disabled for FIPS and crash the running operation.

File	Line	Hash	Usage
agent/context_compressor.py	1269	md5	Content dedup hash (not security)
agent/codex_responses_adapter.py	236	sha1	Function-call ID derivation (not security)

Both are content-hashing use cases, not cryptographic security — usedforsecurity=False is the correct annotation.

Related: Same pattern was previously fixed in #62654 (skills_hub, skills_sync, web_server) and #64062 (qqbot, wecom, yuanbao_media). These two files were missed.

2. assert in production code (P2)

Bare assert statements in production paths are stripped under python -O, leaving silent None dereferences or swallowed errors.

File	Line	Fix
agent/transports/codex_app_server_session.py	399	assert → raise RuntimeError(...)
tui_gateway/server.py	5853	assert → return _err(rid, -32000, ...)

Related: Same pattern was previously fixed in #62659 (7 files, 15 sites). These two were missed.

Testing

# Verify no regressions
python -c "from agent.context_compressor import ContextCompressor"
python -c "from agent.codex_responses_adapter import _derive_responses_function_call_id"

# FIPS simulation (if available)
python -c "import hashlib; hashlib.md5(b'test', usedforsecurity=False)"

Risk

Minimal — each change is a single-line substitution with identical runtime behaviour under normal Python. The usedforsecurity=False flag is a no-op on non-FIPS systems. The runtime guards produce the same error that the asserts would, but survive -O mode.

[teknium1]

Thanks for identifying the two remaining non-security hash callsites; both are still present on current main at agent/codex_responses_adapter.py:236 and agent/context_compressor.py:1578.

Problems

• The new session.activate fallback is unreachable. _sess_nowait() returns (None, error) for a missing session at tui_gateway/server.py:1486-1488, and session.activate returns that error at tui_gateway/server.py:6160-6162 before the replacement guard. Remove the redundant assertion rather than adding a second error contract.
• CodexAppServerSession.compact_thread() retains the same post-ensure_started() assertion at agent/transports/codex_app_server_session.py:661 that this PR replaces in run_turn().
• The PR adds no coverage in the existing target suites, including tests/agent/test_context_compressor.py, tests/agent/test_codex_responses_adapter.py, and tests/agent/transports/test_codex_app_server_session.py.

Suggested changes

• Preserve _sess_nowait()'s existing missing-session response and remove the dead fallback.
• Cover compact_thread() consistently if the assertion-hardening scope remains.
• Add focused FIPS-keyword regression tests.

Automated hermes-sweeper review.

[teknium1]

_sess_nowait() already returns an error whenever session is None, and the preceding if err: return err exits first. This fallback is unreachable; remove the redundant assertion instead of introducing a second missing-session error contract.

[teknium1]

compact_thread() repeats this exact post-ensure_started() assertion at current main agent/transports/codex_app_server_session.py:661. Please apply the same handling there, or centralize the shared invariant, if this PR is intended to harden these production paths under python -O.

[GottZ]

This was generated by AI during triage.

Summary

Twenty-one PRs address or reference this cluster: most annotate non-security MD5/SHA-1 constructors for FIPS compatibility, three change WeCom signature comparison, and two broad SHA-256 sweeps also alter security-sensitive contexts. The diffs range from focused current-path substitutions to overlapping repository-wide batches, while #64808 combines two relevant hash annotations with unrelated and incomplete assertion handling.

Related pull requests

• hash: silence 19 bandit B324 weak-hash warnings (cache keys + protocol fields) #27368 related — (+19/-19) — n/a: Annotates 19 cache, fingerprint, and protocol-required MD5/SHA-1 sites, but the keep_open review requires rebasing the relocated WeCom files, covering current sibling sites, and narrowing the repository-wide scan claim.
• fix(wecom): use constant-time comparison for the callback signature #39049 related — (+7/-1) — n/a: Replaces WeCom's plain signature comparison with hmac.compare_digest in the retired gateway path; despite its keep_open review, the diff is superseded at the live path by fix(wecom): use hmac.compare_digest for webhook signature check #54617 and lacks the requested primitive-level test.
• fix(wecom): constant-time callback signature comparison #43937 related — (+44/-1) — n/a: Adds the same comparison plus spy and None-signature tests, but production and test imports use the retired gateway path; despite its keep_open review, fix(wecom): use hmac.compare_digest for webhook signature check #54617 is the current-path implementation, while this PR's stronger tests remain worth carrying forward.
• fix(skills_hub): pass usedforsecurity=False to md5 cache keys (FIPS compatibility) #48472 related — (+5/-5) — n/a: Adds usedforsecurity=False to the five Skills Hub MD5 cache-key constructors, with the maintainer-bot keep_open verdict identifying a salvage path through a mocked-FIPS regression covering all five paths.
• fix: add usedforsecurity=False to hashlib.md5() calls for FIPS compatibility #51962 [closed] related — (+10/-10) — n/a: Closed reference implementation for non-security MD5 annotations across the compressor, Weixin, WeCom, Skills Hub, and Skills Sync, substantially superseded by the broader fix(agent,tools,gateway,plugins): add usedforsecurity=False to non-security hashlib calls #52783.
• fix: add usedforsecurity=False to hashlib.sha256/sha1 calls for FIPS #51973 [closed] related — (+48/-48) — n/a: Closed broad SHA-256/SHA-1 reference sweep that also marks PKCE, checksum, authentication, and signature-related paths as non-security, illustrating the classification problem repeated by later broad sweeps.
• fix(agent,tools,gateway,plugins): add usedforsecurity=False to non-security hashlib calls #52783 related — (+16/-16) — n/a: Consolidates MD5/SHA-1 annotations across eight files, but its keep_open review requires the omitted Skills Hub cache key, constructor-keyword tests, and removal of the COS authorization digest pending separate protocol verification.
• fix(tools,gateway,agent,plugins): add usedforsecurity=False to hashlib calls #52967 [closed] related — (+18/-18) — n/a: Closed v2 superset of fix(agent,tools,gateway,plugins): add usedforsecurity=False to non-security hashlib calls #52783 with an additional WeCom callback SHA-1 change; contributor discussion identifies it as the same core change and retains fix(agent,tools,gateway,plugins): add usedforsecurity=False to non-security hashlib calls #52783 as the canonical open PR.
• fix(wecom): use hmac.compare_digest for webhook signature check #54617 related — (+32/-1) — n/a: Applies hmac.compare_digest to the live WeCom plugin path, but its keep_open review finds that the added acceptance tests do not enforce the timing-safe primitive and requests a spy-based regression.
• fix: hashlib.md5 FIPS crash in context_compressor.py #56715 related — (+1/-1) — n/a: Adds usedforsecurity=False to the compressor deduplication MD5 call as a focused subset of fix(agent,tools,gateway,plugins): add usedforsecurity=False to non-security hashlib calls #52783, with the keep_open review requesting a mocked-constructor regression or explicit consolidation.
• fix: hashlib FIPS crash in gateway and tools modules #56719 related — (+13/-13) — n/a: Covers QQ Bot, Weixin, Yuanbao, and Skills Hub hashes, but its keep_open review requires separating the COS signing digest, adding FIPS-keyword coverage, and including omitted sibling sites.
• fix: add usedforsecurity=False to hashlib md5/sha1 in skills_hub, skills_sync, web_server #62654 related — (+8/-8) — n/a: Covers five Skills Hub cache keys, Skills Sync directory hashing, and two web-server action-name hashes, with the maintainer-bot keep_open verdict requesting a constructor regression and a corrected complementary-slice description.
• fix: hashlib FIPS crash in qqbot, wecom, yuanbao_media (md5/sha1 without usedforsecurity=False) #64062 related — (+8/-8) — n/a: Covers QQ Bot and WeCom upload checksums plus WeCom callback and COS signing hashes, but its keep_open review requires retaining only confirmed non-security uses and resolving both signature contexts separately.
• fix: hashlib FIPS crash + assert→runtime guards in 4 production files #64808 related — (+6/-4) — n/a: Combines two relevant non-security hash annotations with assertion replacements, but its keep_open review finds the TUI fallback unreachable, identifies the unchanged compact_thread assertion, and requests target-suite coverage.
• fix: hashlib FIPS crash in skills_sync, weixin, web_server (usedforsecurity=False) #65434 related — (+5/-5) — n/a: Covers Weixin checksums, Skills Sync hashing, and web-server action names, but the maintainer-bot keep_open verdict requires adding the five omitted Skills Hub cache keys and a mocked-FIPS regression.
• fix: add usedforsecurity=False to hashlib calls in agent/ (FIPS compliance) #65561 related — (+11/-10) — n/a: Broadly marks agent-side SHA-256 uses including PKCE and checksum paths with usedforsecurity=False, without separating security contexts from cache or deduplication uses or adding restricted-runtime tests.
• fix: add usedforsecurity=False to hashlib in tools/ and plugins/ (FIPS compliance) #65570 related — (+20/-19) — n/a: Broadly annotates tools and plugin SHA-256 uses including Feishu signature verification and PKCE, while its keep_open review requires removing security-context changes, narrowing the scope, and adding focused restricted-hash tests.
• fix: add usedforsecurity=False to hashlib.md5 in weixin adapter (FIPS crash) #66857 related — (+2/-2) — n/a: Precisely annotates the two Weixin MD5 deduplication and upload sites, with the maintainer-bot keep_open verdict confirming those exact substitutions against current main.
• fix: add usedforsecurity=False to hashlib.sha1 in tui_gateway/server.py #72370 related — (+1/-1) — n/a: Annotates the TUI MCP revision SHA-1 constructor, with its keep_open review correcting the symptom to loss of revision-based reload gating and requiring a real-helper regression.
• fix: add usedforsecurity=False to hashlib.sha1 in tui_gateway/server (FIPS crash) #73278 [closed] related — (+1/-1) — n/a: Closed duplicate of fix: add usedforsecurity=False to hashlib.sha1 in tui_gateway/server.py #72370 with the identical MCP revision-hash change, closed by contributor triage for missing regression coverage.
• fix: add usedforsecurity=False to hashlib.md5 in context_compressor (#FIPS) #74632 related — (+1/-1) — n/a: Annotates the current compressor deduplication MD5 site reached by normal compression and proactive pruning, with contributor reviews requiring a focused duplicate-output regression using a rejecting MD5 stub.

Duplicates

#52967 is the closed v2 duplicate/superset of #52783, and #73278 is the closed identical duplicate of #72370; #39049 and #43937 overlap #54617 but target the retired path. #48472 is the Skills Hub subset of #62654, #56715 is a focused subset of #52783, and #66857 duplicates the Weixin slice present in #56719 and #65434.

Suggested consolidation

Keep #64808 open with a salvage path: retain its two hash annotations, remove the unreachable TUI fallback, cover the sibling compact_thread assertion consistently, and add focused constructor and session tests. After carrying #43937's spy and None-handling coverage into live-path #54617, close #39049 and #43937 as superseded duplicates despite their keep_open reviews because their diffs edit the retired gateway path; keep the remaining focused current-path PRs open only for their documented tests or unique sites, and close overlapping subsets once those sites are explicitly carried into a tested consolidation.

Cross-PR triage: Reviewed 21 pull requests and 0 issues in this complex. Diffs were read for 20 of 21 PRs (rest unavailable); Assessment working set: 112 kB of PR diffs, 28 kB of issue/PR text, 22 kB of discussion (31 comments), 0 verify verdicts. verdicts reflect diff content, not PR titles. Part of an automated triage batch.