← graph

remove-validation-cli

Remove --validation CLI flag and all related hard-gate text from CLI/prompts/docs

Metadata

Statusdone
Assignedagent-92
Agent identityf51439356729d112a6c404803d88015d5b44832c6c584c62b96732b63c2b0c7e
Created2026-04-26T13:03:27.081375627+00:00
Started2026-04-26T19:43:08.589328667+00:00
Completed2026-04-26T20:04:51.331372524+00:00
Tagseval-scheduled
Eval score0.05
└ blocking impact0.05
└ completeness0.00
└ coordination overhead0.15
└ correctness0.00
└ downstream usability0.00
└ efficiency0.10
└ intent fidelity0.04
└ style adherence0.00

Description

Description

User decision: hard validation gates always get fucked up; agency evaluator (auto_evaluate + FLIP) handles validation by reading the ## Validation section of task descriptions. Remove the --validation CLI flag and all references that suggest using it.

Files to clean (from grep -rn '--validation' src/)

  • src/cli.rs:237,241,453 — --validation flag definition on add and edit commands. Remove the flag entirely. If feasible behind a deprecation cycle, keep accepting it for one release with a warning that it is a no-op.
  • src/config.rs:2715 — comment 'New tasks should use --validation=llm instead.' — remove or rewrite to point at the agency-evaluator pattern.
  • src/service/executor.rs:239,242,408,487,490,723,760,763,2871-2874 — agent-prompt text + tests asserting the prompt mentions --validation. Replace with: 'put validation criteria in the ## Validation section of the task description; the agency evaluator scores against it.'
  • src/commands/edit.rs:269 — error message that suggests --validation=llm. Drop the suggestion.
  • src/commands/quickstart.rs:212,214 — VALIDATION section in quickstart output. Rewrite to describe the agency evaluator path with no flag.
  • wg add runtime hint that prints '2. Use --validation=llm for an LLM verification gate at completion time' (likely in the same area). Remove option 2 entirely; only option 1 (## Validation in -d) remains.

What stays

  • PendingValidation task status itself stays — it can still be entered via the agency-driven path (low FLIP score, eval gate, manual flag).
  • wg approve / wg reject stay — the human-review path off PendingValidation is unchanged.
  • Anything in the agency evaluator code that scores ## Validation sections stays.

Out of scope

  • Removing PendingValidation as a status (not asked).
  • Removing legacy --verify (already removed; only doc text remained and was cleaned in a prior turn).

Validation

  • Failing tests first:
    • test_cli_add_no_validation_flag — wg add --help no longer mentions --validation; passing --validation=llm errors with 'unknown flag' OR (if deprecation kept) prints a one-line warning and is otherwise a no-op
    • test_quickstart_no_validation_flag — wg quickstart output mentions ## Validation section but not the --validation flag
    • test_executor_prompt_no_validation_flag — agent prompts assembled in build_prompt (or wherever) do not contain the string '--validation'
  • Implementation removes the flag and rewrites prompt/quickstart text per spec
  • cargo build + cargo test pass with no regressions
  • Manual smoke: wg add 'test' --validation=llm either errors cleanly or warns + ignores; wg quickstart output describes only the agency-evaluator path; wg add 'test' followed by inspecting the dispatched agent's prompt shows no --validation mention

Depends on

Required by

Log