Batch A: independent bug fixes and security hardening

[DavidJBianco]

Summary

Cherry-picks five independent, non-overlapping PRs from the Codex/Aardvark batch onto dev:

• fix(mcp): correct hypothesizer argument mapping in MCP wrapper #78 fix(mcp): correct hypothesizer positional argument order (was silently swapping local_context and local_data_document)
• fix: reset data discovery tab session state on hypothesis change #79 fix: reset Data Discovery tab session state using correct keys (Discovery_* not Data Sources_*)
• Sanitize branch-derived Docker tag in Makefile targets #80 Sanitize Makefile Docker tag derived from git branch name (prevents tag injection via crafted branch names)
• fix: guard stdio timeout parsing against invalid config values #81 fix: guard stdio timeout parsing against invalid/null config values (fallback to 30s default)
• fix: mitigate SSRF risk in report URL validity checks #82 fix: block internal/private URL targets in eval URL validity checker; disable redirect following (SSRF mitigation)

Notes

• fix: mitigate SSRF risk in report URL validity checks #82 blocks IP-literal private targets and disables redirect-following. Residual gap: public hostnames that DNS-resolve to private IPs are not blocked — acceptable for an offline eval tool, noted as a review comment on the upstream PR.
• All five cherry-picks applied cleanly onto dev; python3 -m compileall peak_assistant passes.

Test plan

• Run python3 -m compileall peak_assistant — should pass
• Run PYTHONPATH=. pytest -q tests/unit_tests/test_mcp_stdio_timeout.py in a full env to verify timeout fallback
• Smoke-test MCP hypothesizer tool to confirm correct hypothesis output (validates fix(mcp): correct hypothesizer argument mapping in MCP wrapper #78)
• Verify Data Discovery tab resets on hypothesis change (validates fix: reset data discovery tab session state on hypothesis change #79)

🤖 Generated with Claude Code

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 05b4499542

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Count blocked URLs in validity score denominator

The URL validity score now samples only safe_urls, so private/internal links are excluded from both sample_size and invalid_urls. In reports that mix blocked links with a few reachable public links, this can produce an artificially high URL-validity score (e.g., 1 valid public URL + many blocked URLs can still score near 100), which skews evaluation results. To preserve metric integrity while keeping SSRF protections, blocked URLs should still contribute to the invalid/denominator side of scoring.

Useful? React with 👍 / 👎.