# Verification level: how much independent checking a skill runs One setting, `verification`, decides whether a skill's verifier-subagent steps run, and how many. Each plugin keeps it in its own team file; this page owns what the levels mean. It never touches CI merge gates: the pull request lanes (`pr-verify`, the review lanes, `ci-status`) run whatever the level says. ## Keys In a plugin's team file, `/.yaml`, validated by that plugin's schema. `` is the convention home named by the pointer line, read at the same default-branch commit by `lib/resolve-convention-home.sh`; with no pointer line, or a declined one, it is `docs/conventions`. A broken pointer skips the layer. ```yaml verification: light # the plugin's level verification_skills: confirm: strict # one skill's level; beats the plugin's ``` | Level | What an adopting skill does | |---|---| | `off` | everything `light` does, and also skips the skill's own real-check steps that its text names as `off`-skippable | | `light` | the default: skips the verifier dispatch and writes `verification: skipped (light)` where the verdict would go; it never checks inline in its place | | `standard` | runs the verifier dispatches the skill's text describes, its optional passes left optional | | `strict` | always runs the verifier passes that `standard` leaves optional, such as a carve-out for mechanical, behavior-preserving changes | Where the verifier is the product (a scan whose findings are the verifier's output), `light` emits the results labeled unverified instead of skipping them. ## Resolution `read-verification.sh --plugin --skill `, shipped in each adopting plugin's `lib/` (source `lib/read-verification.sh`), resolves the level and always prints one line, for example `verification: light (docs/conventions/implementation.yaml verification at origin/main, commit )`. - `verification_skills.` wins; otherwise `verification`; otherwise `light`. - To raise the level, set `verification: standard` (or `strict`) in the team file on the default branch, or `verification_skills.` for one skill. - A plugin whose team file sits at a fixed path whatever the convention home passes it with `--file `, and the convention home is not read: `bugs`, `discipline`, `discovery`, `docs-hygiene`, `education`, `implementation`, `mutation-testing`, `planning`, `playbooks`, `session-flow`, `testing`, `verification` and `work-items` pass `--file docs/conventions/.yaml`. Without `--file` the file is `/.yaml`. - **Team layer only, read from the default branch.** Floor direction `team-only` ([config-cascade](../config-cascade/README.md#sanctioned-exception-class-policy-floor-precedence-inversion)): the file is read at `origin/` after a fetch, never the working tree, so a branch cannot lower its own level. There is no user option and no personal override. When the working tree's copy gives another level, a second line names both copies. - The root rule of `lib/config-root.sh classify` comes first: a `$HOME` or non-repository root skips the layer. So do no `origin`, a default branch name outside letters, digits, `.`, `_`, `/` and `-` (or one with a leading `-` or `..`), an unfetched default branch, and a broken convention-home pointer. Each skip resolves `light` with its reason, as do no file and a file that sets no level. - An invalid value at either key (not `off`, `light`, `standard` or `strict`, or a list or map), and a file that does not parse, are named and resolve `standard`: a team that set a level keeps a floor through a typo. A failed fetch reads the last fetched copy and says it may be stale. ## Interactions with other settings - **`proof_level`** (verification plugin, stricter wins) is unchanged: `off` never drops evidence a resolved `live` or `strict` proof level requires. - **`verify_mechanical_phases`** (implementation plugin, floor `true`): `true` keeps the phase verifier on every phase under `light` and `off`. `strict` implies `true`. - **`pre_ready`** (source-control) decides whether the ready flow's verify step runs; `verification` decides only what happens inside it. - **The PR-prep gate.** When PR prep calls `/verification:confirm`, a `NOT VERIFIED` whose only cause is `verification: skipped ()`, from the level resolved at the default branch (a set key or the default), does not block PR creation: the report names it and the PR body carries the skipped line. A level that turns verification down never becomes a new blocker. Any other `NOT VERIFIED` cause still blocks, and Stage 1's mechanical checks still run and gate under `light` and, in `/verification:confirm`, under `off`. ## Exempt skills The level governs a pass that checks the skill's own output; a pass that produces the skill's deliverable is not governed. These skills do not read the level: `/planning:audit-answers`, `/performance:verify`, `/discipline:scrutinize-dont-coast`, `/review:quality-gate`, `/review:fanout`, `/planning:devils-advocate`, `/review:security-review` and `/planning:interview`. Each is itself the independent check when invoked, except `/planning:interview`, whose contradiction pass, assumption sweep and quote check produce the questions it exists to ask. `/coupling:reduce` and the conventions lens of `/architecture:improve` do not read it either: reproducing a scan finding produces the finding they apply or propose for recording, so it runs on every run. ## Adopters | Skill | Steps the level governs | |---|---| | `/verification:confirm` | the fresh-context verifier for the Stage 2 verdict ("Independence of the verdict") | | `/implementation:implement-dispatch` | the phase verifier and the post-phase source commit verifier (Gates, Phase boundaries) | | `/discovery:research`, `/discovery:explore`, `/discovery:trace-intent` | the sibling verifier (parent contract, "The sibling verifier, stated once"); `strict` widens research's grading to every row and claim | | every `/discipline` corrector on the shared method but `/discipline:scrutinize-dont-coast`, and `/discipline:sweep-all` | the fresh-context confirmation of a clean audit on load-bearing work (shared method, step 2), the fresh-context re-derivation (step 3) and the fan-out tiers' dispatch; `strict` takes do-your-research's fan-out tier at `full` depth | | `/education:quiz-me` | the fresh-context answer-key pass | | `/planning:plan` | the plan-reviewer dispatch in Step 3; `strict` dispatches it for every plan, which `plan_stress_test`'s `auto` otherwise skips on LOW plans | | `/session-flow:orchestrate`, `/session-flow:workflow`, `/session-flow:retro` | imperative 3's verifier in a primed session; the fresh-context reviewer in pre-PR step 2, without which Stage 10 is not done; retro's single-pass analysis and scoring under `retro_lenses: 1` | | `/harness-config:audit-automation-gaps` | the fresh-context refusal gate (2.3), whose verdicts `light` lists as `unverified`, and the Phase 4 step 6 reviewer | | `/harness-config:audit-instructions` | the Phase C verifiers (per lane, B2 and I33) | | `/harness-config:audit-pass` | the Phase 5 apply-verify dispatch and the Phase 6 main-session verification | | `/harness-config:audit-prompting-postures` | the Phase D verifier, on proposals and on verdicts | | `/bugs:scan` | the Step 4 gate, the product: `light` reports candidates as `unverified` | | `/mutation-testing:audit` | the Phase 4 triage, the product: `light` reports survivors as unclassified | | `/testing:cleanup` | the Cannot-fail-only mode's step 2 skeptic: `light` clears no row, so that mode deletes nothing | | `/testing:write` | the Handoff advisor checkpoint for a HIGH or CRITICAL suite; `strict` calls it for every suite | | `/debugging:debug` | the fresh-context outcome verifier when no outcome-verification capability is used | | `/implementation:implement` | the fresh-context verifier at each phase boundary (item 1 of the ritual) and the Step 5 advisor checkpoint; `strict` calls the advisor for every change | | `/docs-hygiene:compress` | the semantic-diff subagent and its revert pass (Hard rules); `compare` is not governed | | `/docs-hygiene:audit-derivability` | the empirical spot-test and the keep-sample (Classification method) | | `/docs-hygiene:audit-noise` | the fresh-context verification pass over the merged verdicts | | `/docs-hygiene:extract-ssot` | the adversarial-review round for a `WARN-borderline` cluster | | `/code-tidying:batch-simplify` | the per-group refutation verifier (Phase 6, "The verification level") | | `/ai-slop:audit` | the fix flow's semantic-diff subagent | | `/attribution:audit` | the review pass, the refutation pass, and the fix flow's semantic-diff subagent; not the judges | | `/explainer-video:produce` | the step 7 fresh-context frame judge on the delivery render: `light` reports the frames as not independently judged | | `/knowledge:video-digest` | the 9.4 vision fidelity spot-check: `light` reports the slice structural complete only | | `/knowledge:docpage-digest` | the Phase 4 verifier arms on the single-page path, the product: `light` labels the slice unverified; the batch workflow runs both arms at every level | | `/event-storming:simulation` | the fresh-context rubric scorer ("How to Run an Evaluation", step 3), the product: `light` leaves every criterion `unscored` | | `/architecture:improve` | the deepening lens's Phase 1.5 reproduction, the product: `light` presents those claims `unverified`; the conventions lens is not governed | | `/multi-agent:audit-defaults` | the `drift-audit` workflow's skeptic panel, the product: `light` reports every finding `unverified` and proposes no diff | | `/playbooks:fable-5` | the fresh-context verifier gate (orchestration chapter, "Fresh-context verification"); the self-review floor runs at every level, and `strict` drops the mechanical-batch exemption | | `/plugin-quality:audit` | the adversarial re-examination seam, the product: `light` labels the findings `unverified`; `standard` keeps its `high`-effort gate and `strict` runs it at every effort | | `/songwriting:co-write`, `/songwriting:diagnose` | the hard gate's skeptic refutation row; `standard` keeps it skippable by the writer's named reason, and `strict` runs it on every batch; the rubric's counted passes run at every level | | `/work-items:work` | the bug path's step 2 capture check, a fresh-context subagent per reproduction capture; `light` leaves the runs unconfirmed | | `/codebase-health:audit` | the Phase 2 separate false-positive validator; `strict` dispatches it after an inline Phase 1 too, and `light` reports the findings `unverified` | | `/developer-experience:build-cli` | the review mode's fresh-context reviewer, the product: `light` runs the review in the main session and labels every gap `unverified`; `strict` dispatches the reviewer for every review | | `/harness-ops:audit-install-state` | the fresh-context cross-review (evidence-discipline section 5); `strict` dispatches it for a single-agent run too | | `/harness-ops:changelog` | the local verifier of repo-side facts, in `diff` and in `apply`'s re-check of saved rows: `light` marks those rows `unverified` | | `/playwright:demo-video` | the step 8 independent frame review, items 1-6 (quality); item 7, the secrets check that gates posting, runs at every level | | `/writing:be-concise` | the fresh-context semantic diff; `strict` runs it on every rewrite, whatever its size | | `/wizard:generate` | the Phase 3 step 2 fresh-context read of the script, non-secrets items only; the library diff against `template.sh` and the destination and secret-name items run at every level |