fix(pi): expose owning Taskflow run identity to children

[heggria]

Summary

Fixes #172.

• Add optional RunOptions.runId and populate it from the owning state in the imperative path and both event-kernel call builders.
• Export it as PI_SUBAGENT_RUN_ID in Pi children, replacing a stale inherited Pi marker without mutating the parent environment.
• Preserve direct-call behavior when no run ID is supplied, existing DEVLOOPS_RUN_ID ownership, child isolation, and other host adapters.
• Add 17 regression tests covering seven agent-call phase families in both engines, nested owning identities, flow/phase/argument decoys, and the actual child environment.

Verification

• Reproduced six failing identity assertions before the patch using the actual Taskflow store/runtime/Pi runner and spawned local probe processes.
• Full workspace unit suite: 2,418 passed, 0 failed.
• pnpm run typecheck: passed.
• pnpm run build: passed.
• Real-process integration harness with unmodified current dev-loops async-start guard: 13 passed.
• Independent process checks against current and filing-time dev-loops guards: 9 passed.
• git diff --check: passed.

Tests ran on Linux with Node 24.19.0 and pnpm 11.19.0. A writable isolated HOME was used for workspace-control storage. Integration probes replaced only the Pi CLI executable; no live Pi/model end-to-end execution or external GitHub workflow was performed locally.

Independent main landing (2026-10-07)

This fix targets main directly and does not depend on release PR #142 or the dependency upgrades in #174.

Scanner diagnosis and minimal adjustment

• Re-scanned current main 48a5ea3 and the run-identity tree with the same plugin-scanner 2.0.1116 / cisco-ai-skill-scanner 2.0.14, balanced policy, and completed static/bytecode/pipeline coverage. Both scored 78/100; the complete finding lists were identical.
• The diagnostic exact-head scan exposes both highs as UNICODE_OBFUSCATED_INSTRUCTION, in packages/codex-taskflow/plugin/skills/taskflow/SKILL.md:3 and configuration.md:1. Both documents were byte-identical to main.
• Inspected the installed 2.0.14 rule: ordinary ellipses trigger NFKC normalization, while an unrelated document-wide regex matches text such as "Read-only ... zero tokens". Its location is the first non-ASCII line, not a malicious payload location. These are reproducible documentation false positives, not run-identity defects.
• Cisco 2.0.12 used by fix(ci): consolidate dependency upgrades and retain deep scans #174 does not contain this rule. Its green scan is therefore not evidence that these findings were remediated.
• Normalize only typographic ellipses (… to ...) in skills-src/taskflow/core.md and configuration.md, then regenerate the six host copies. No instructions, permissions, runtime behavior, scanner versions, rules, thresholds, or suppression baselines changed.
• Keep full JSON findings visible in CI, including failures.

Verification for final head

Head: f6094917507b79e0d1a610a08d03f0a035924d66.

• Local run-identity regressions: 17/17 passed, including real child-process environment capture.
• Skill build/drift/host filtering tests: 3/3 passed; build-skills --check and git diff --check passed.
• Final local scan on the unchanged scanner versions: 81/100, critical 0, high 0, medium 5, low 9, info 5. Cisco skill scanning completed with balanced static/bytecode/pipeline coverage. Optional Cisco MCP scanning remains unavailable, as before.
• Remote full CI and Plugin Security Scan both passed on this exact head. All 10 CI jobs succeeded, including Node 22/24 typechecks and full unit suites, three-OS process supervision, package builds/consumers, MCP E2E, website export, and CodeQL. The remote scan confirmed 81/100, critical 0, high 0, unchanged scanner/Cisco versions, completed balanced static/bytecode/pipeline coverage, and zero skipped targets.

The pre-adjustment diagnostic head 623a21d also passed all 10 CI jobs. Historical local verification above is retained as historical evidence, not claimed to have been rerun in full locally today.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-07T14:09:51.397930Z	f609491	Draft marked ready

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" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.