[mashkovd]

Why

The human-input producer shipped in mctl-agents#558 (release 1.66.0). After a human answers a sealed question, the DevLoop continuation sends the accepted answers as the human_input_responses parameter (dev_loop.py), and the investigator reads them as --human-input-responses. mctl-agents-investigate did not declare the parameter, so the answer never reached the run. As docs/runbooks/human-input-e2e.md in mctl-agents notes, the continuation then re-asks the question and the workflow skips it by hash. This is the gitops half of mctl-api#372 item 3.

What

cwft-mctl-agents-investigate.yaml:

• Declares the optional human_input_responses parameter, empty by default.
• Adds the env WORKFLOW_HUMAN_INPUT_RESPONSES.
• Appends --human-input-responses "$WORKFLOW_HUMAN_INPUT_RESPONSES" only when the value is non-empty. Every run without answers keeps its exact current argv.
• Reads the value through env, never spliced into the shell, the same as issue_url and the correlation ids.

Order and what this does not do

• This must land before mctl-api declares the parameter in registry.go (it is in devLoopParamsPendingDeclaration today). Until then mctl-api strips it, so this PR alone changes no live run.
• The flag requires an issue-investigator release with mctl-agents#558 (1.66.0+). Releases before that exit non-zero on it. Only continuations send it, and only once mctl-api forwards it.
• The capability stays inert while the template runs the legacy resolver mode (agent-platform README, "Kill switch"). This PR does not change the resolver mode.

Refs mctlhq/mctl-agents#473, mctlhq/mctl-api#372

🤖 Generated with Claude Code

https://claude.ai/code/session_012kWiYf1BX6wcXB2kc8x15U

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[claude]

Claude finished @mashkovd's task in 2m 31s —— View job

---

Reviewing PR #1581

• Gather context (diff, CLAUDE.md conventions)
• Analyze cwft-mctl-agents-investigate.yaml changes
• Post findings / verdict

Verdict: Changes requested (1 P2, 0 P1, 0 P3)

The implementation itself follows the file's established pattern correctly:

• New human_input_responses parameter defaults to "" — zero behaviour change for every existing caller.
• Read through WORKFLOW_HUMAN_INPUT_RESPONSES env, never shell-interpolated directly — the safe pattern enforced by scripts/validate-shell-param-interpolation.py for every sibling parameter (work_item_id, execution_id, temporal_workflow_id, temporal_run_id, execution_request_id).
• Appended to argv only when non-empty, in the same run-investigator container whose env defines it.

P2 — missing test coverage: tests/test_cwft_investigate_work_context_argv.py exists specifically to pin this exact set -- argv-building block for every optional identifier added so far — omit-empty passthrough (T1), flag ordering (T2/T6/T7), and shell-injection safety (T4/T8). This PR extends the same block with --human-input-responses (lines 414-415) but doesn't extend that test, so a future change to the block can silently regress the new flag with nothing to catch it. Posted inline with a suggested fix (add WORKFLOW_HUMAN_INPUT_RESPONSES to the test's run() env dict and mirror T2/T3/T6/T8 checks for the new flag).

[github-actions]

Antigravity (agy) review — informational

Model: gemini-3.8-flash-medium · commit bba0419.

Findings

[P3] No test visible in this diff for workflow parameter forwarding

• File: platform-gitops/argo-workflows/cluster-templates/cwft-mctl-agents-investigate.yaml:123
• Anchor: + - name: human_input_responses
• Concrete failure scenario: The workflow template adds parameter human_input_responses and conditional forwarding logic to append --human-input-responses when the environment variable is populated. There is no test visible in this diff under tests/ to validate template schema compliance, parameter default value handling, and command argument synthesis when human_input_responses is provided versus when omitted.
• Proposed fix: Add a test under tests/ that parses cwft-mctl-agents-investigate.yaml to verify schema validation and test shell argument assembly for both empty and non-empty values of WORKFLOW_HUMAN_INPUT_RESPONSES.

[claude]

Has P1/P2 findings, changes requested: 1 P2 (0 P1, 0 P3). The implementation itself is correct and follows the file's established pattern exactly: optional parameter defaulting to empty (zero behaviour change when omitted), env-bound (never shell-interpolated — consistent with scripts/validate-shell-param-interpolation.py's safe pattern), appended to argv only when non-empty, and the env var lives in the same container as the script that reads it. Remaining gap is missing test coverage: tests/test_cwft_investigate_work_context_argv.py exists specifically to pin this exact argv-building block for every prior optional identifier (work_item_id, execution_id, temporal_workflow_id, temporal_run_id, execution_request_id) — omit-empty passthrough, flag ordering, and injection safety — and was not extended for the new human_input_responses flag.

[claude]

Claude finished @mashkovd's task in 1m 31s —— View job

---

Reviewing PR #1581 (commit afb2587d)

• Gather context (git diff 95710e1...afb2587, CLAUDE.md conventions)
• Analyze cwft-mctl-agents-investigate.yaml changes
• Analyze tests/test_cwft_investigate_work_context_argv.py changes
• Cross-check CI (Validate Manifests, yamllint, PR feat(investigate): forward human_input_responses to the investigator #1581 test run all green on this commit)
• Post findings / verdict

Verdict: Approved (0 P1, 0 P2, 0 P3)

This commit (afb2587d) extends the prior commit (bba0419e) with exactly the test coverage flagged as missing in the earlier review round:

• run()'s env dict now sets WORKFLOW_HUMAN_INPUT_RESPONSES alongside the other five identifiers.
• T9 pins omit-empty passthrough and that --human-input-responses is appended last, after every other optional flag, both alone and combined with all five siblings.
• T10 proves injection-safety: a hostile value containing ;, `, and $(...) survives as a single argv token via set -- and never executes (sentinel-file check), the same guarantee T4/T8 establish for the other identifiers.

Verified independently:

• human_input_responses is bound through env: (WORKFLOW_HUMAN_INPUT_RESPONSES → {{workflow.parameters.human_input_responses}}) and read as "$WORKFLOW_HUMAN_INPUT_RESPONSES", never interpolated directly into script text — scripts/validate-shell-param-interpolation.py's shell_blocks() only scans script.source/command/args, and this value never appears there, so no BASELINE entry is needed.
• The set --/if [ -n ... ] append block matches the existing five-flag pattern exactly: default "", omit-when-empty, no invented value.
• CI on this commit is green: Validate Manifests, yamllint, and the PR #1581 test-suite run (which includes this test file) all passed.

No further changes requested.

[github-actions]

Antigravity (agy) review — informational

Model: gemini-3.8-flash-medium · commit afb2587.

No significant issues found.

[claude]

No P1/P2 findings (0 P3). Commit afb2587 adds the test coverage (T9/T10) the prior review round asked for — omit-empty passthrough, flag ordering, and injection-safety for human_input_responses, mirroring T2/T4/T6/T8. Good to merge.