Justify

Repository context. Gather first

Collect this with an individual Bash call, never combined into a single invocation with anything else:

• Branch, git symbolic-ref --quiet --short HEAD

Treat a failure (not a repository, git unavailable) as an unknown value and carry on. Keep it as a separate body Bash call rather than a pre-compute line: the harness runs a skill's whole pre-compute block as one shell invocation, and a worktree-isolated session refuses a compound command that contains git. The call prints the branch name on stdout and fails with no output on a detached checkout rather than printing a sentinel, so read its exit status to tell those two apart rather than matching on a string, and take the branch name itself from stdout.

Purpose

Point at one artifact and ask whether it has earned its place. The two-part test is the operator's: a reason existed when it was built, and that reason still holds today. The method computes both already, so this skill supplies a lane rather than a second method.

Neither the method nor the lane is restated here. Read ${CLAUDE_PLUGIN_ROOT}/context/scrutiny-method.md before judging anything and cite its sections in the findings; every bare §N in this file is one of its sections. Then read ${CLAUDE_PLUGIN_ROOT}/context/justification-lane.md before the first target: it carries this lane's item inventory, its five layers with their probes, its evidence sources on the §2 tiers, its protected classes, the routing precedence, and its known limits. Its sections are numbered, and this file refers to them as "lane section N".

The verdict is the deliverable, and the conversation after it is the point. Report first, then discuss. The remedy is refactor or remove, decided by the operator, and it is carried out by the owner the finding names, never here.

Read-only contract

${CLAUDE_PLUGIN_ROOT}/skills/audit/SKILL.md, section "Read-only contract", governs, and is not restated. Five things are specific to this lane:

• The only write that is this lane's own is the findings artifact, at the memory-tier home resolved through ${CLAUDE_PLUGIN_ROOT}/reference/topic-docs.md. Never the artifact being judged. Emit the read-only opening line immediately after resolving the home, naming that home as where a finding would be written rather than asserting one was: this lane may end up filing no row, and a line that named the path as written would then be a false statement the operator has no reason to doubt. The two auxiliary writes the governing contract sanctions, the topic-docs self-ignore guard and the concern-file persistence on the resolution rungs, are unaffected and still happen: they belong to the binding this lane runs, not to this lane, and skipping the guard would leave the memory root un-gitignored.
• Always mode: targeted, with targets naming what this run examined. A run of this lane never writes mode: walk, because it never walks.
• The frontmatter this lane writes is type: overengineering-findings, schema: 2, mode: targeted, targets, this run's own date and branch, and a scope carrying the prior artifact's value forward with these targets' layers added. type is named first because a first pointed run that writes a row at a home with no artifact creates the file, and it is the selector every consumer matches on: an artifact written without it is one no consumer finds. An on-disk schema of neither 1 nor 2 stops the run with a visible message, because an unrecognized shape cannot be merged into without guessing; a schema: 1 artifact is merged into and rewritten at 2.
• Re-read before write. Load the on-disk artifact immediately before writing, merge against that copy, and read its date to see whether another producer wrote while this run was working. The merge rules protect only what the writer actually read.
• A run that wrote no row writes no artifact, however it got there. The condition is the row count and nothing else, so do not read the examples below as the only ways to reach it. A run reaches it when the no-target ladder ended at an offer or a question, when every target routed away, when every target was declined as an ambiguous heading, or in any other way that leaves the run with nothing filed. Each emits its inline report in full, naming the rung, the routing, or the collision, and none persists anything. Writing an artifact for such a run would stamp a new date and targets on the file and recompute its summary on behalf of a pass that judged nothing, over a walk that did. The case below is different: a detached checkout declines the write for a run that did judge something.

A detached checkout has no branch identity

${CLAUDE_PLUGIN_ROOT}/skills/audit/SKILL.md, section "A detached checkout has no branch identity", governs verbatim: prefer a logical ref the environment supplies, after normalizing and validating it as a branch name; otherwise persist nothing and say why, in these words:

detached checkout, no logical ref supplied; no branch identity, so no findings artifact is written

The pass still runs and the inline report is still emitted in full. What is declined is the persisted write, not the judgment.

Arguments

Parse $ARGUMENTS as one target. Three forms are judged as given, and two more are accepted and widened before judging:

An ambiguous heading is refused, never disambiguated. Where the full ancestry of a path#heading target is still not unique within the file, decline the target, name the collision, and ask for a whole-file target or a disambiguating rename. Never fall back to the first occurrence or to an occurrence count: both are positional ordinals under another name, and the next edit renumbers them, so the same item would derive two ids across two runs. The rule is the contract's, in ${CLAUDE_PLUGIN_ROOT}/context/findings-artifact.md, section "Finding ids".

A line or comment target widens, per lane section 2. Widen to the enclosing heading where the file has headings, to the file otherwise, and state the widening in the report's first line as a rule with its reason: what was pointed at, what is being judged instead, and that an identity resting on a line number derives a different id as soon as an edit above it moves the line. Where the target was a comment, name code-tidying:dissolve-comments as the owner of comments in that same line.

No target: the ladder in lane section 10, and it never sweeps. State which rung was used.

1. Conversation context. Where the session has been discussing an artifact, infer it and confirm the inference in one line before walking.

2. Offer git-age discovery, and wait. Describe what it would rank, run it only on an answer:

   git log --diff-filter=A --name-only --format='%ad' --date=short -- . | awk 'NF'
   

   Post-process into (path, first-seen date) pairs, oldest first. The output is a candidate list to choose from, never a finding set.

   Corroborate before presenting, per lane section 10. Age is not evidence of disuse. Check each ranked path for inbound references under the varied query forms of lane section 7, counting a citation from anywhere in the repository including the path's own directory, and drop the ones that have any. Present what survives with both counts, the ranked and the surviving, so the operator can tell an uncited path from an unsearched one.

3. Ask what to point at.

Before the walk

1. Resolve the branch identity, then the artifact home, exactly as ${CLAUDE_PLUGIN_ROOT}/skills/audit/SKILL.md, section "Before the walk", step 1 does. The branch call in "Repository context" above yields a branch name on stdout, or fails with no output (detached HEAD or no checkout). Read its exit status to decide whether the lookup succeeded, then take the identity from stdout: the status answers only whether there is a branch, and the name itself is the output. Never infer an identity from a failed call, and never accept the literal HEAD as one. Run the topic-docs binding's whole rung order rather than assuming the default's shape.
2. Run the shared preflight, ${CLAUDE_PLUGIN_ROOT}/skills/audit/context/surface-walk.md, section "Preflight". Its sanctioning-record probe matters here: a repetition a record sanctions and a check maintains is never duplication to collapse.
3. Add the lane's four preflight items, lane section 7: vary the query form before any absence claim, make a retirement cost more than a keep, write targeted mode, and re-read before write.
4. Apply the routing test, lane section 1, before classifying anything. A target an enforcement probe inventories whole routes to /overengineering:audit and writes no row; a target only partly inventoried is classified with the routed part named in Routed-to.
5. Resolve consumer configuration and load the prior artifact, as the sibling does. This skill writes Status: OPEN on a finding it has not seen and carries every other status forward. Surface any verdict that moved under a carried-forward judgment, per merge rule 5: a direction flip, and equally a same-direction change to what the acceptance authorized. This lane is the second producer, the flag is written only on merge by whichever producer recomputed the row, and no consumer re-derives it, so a move this step fails to flag is a move nobody surfaces.

The walk

One target is one item, unless the container rule in ${CLAUDE_PLUGIN_ROOT}/skills/audit/context/surface-walk.md, section "Granularity", splits it into mechanically-listed members. The per-item loop is that same document's, section "The per-layer loop", applied once rather than per layer: identify, classify, answer the three liveness questions independently (§3), reconstruct intent (§4), rediscover (§5), weigh cost (§1 and §6), verdict (§6), owner (§12), write.

Verdicts and evidence

Every verdict is one of the six tokens in §6, argued in carry cost (§1), and cites at least one empirical source or is UNPROVEN naming the tier consulted and whether it was silent or unavailable (§2). Beyond that, every row this lane writes carries:

• Basis, per ${CLAUDE_PLUGIN_ROOT}/context/findings-artifact.md, section "Per-finding fields". A row that measured nothing and rests on nothing else, whether no tier was consulted or every consult came back silent or unavailable, is UNPROVEN and unexamined; where the verdict rests on the non-derivable oracle or a protected-class match it is class-inferred instead, per lane section 9. A KEEP is measured or it is not a KEEP.
• Ablation: n/a, spelled with the contract's own capitalisation, because that gate does not apply on this lane's layers (lane section 8). The earned-keep gate is the one every row answers, and a row fusing a class claim with an earned-keep verdict is a defect, not a shortcut. The check constituent of every id this lane derives carries justify as its producer segment.

Intent checkpoints

V1 is attended only. Report first, then open the discussion.

Where intent reconstruction scores MEDIUM or LOW (§4), or where tiers 1 to 3 are silent and tier 4 could exist, ask before writing UNPROVEN. Recommendation first, one small numbered set of questions, batched rather than drip-fed. Reuse planning:interview's question mechanics where it is installed and ask the same questions inline otherwise; the questions are what matter.

"I don't know" is an accepted answer. It records UNPROVEN with the tier named, and resolves the item in neither direction.

Boundary

This lane reports and hands off. Each route is presence-gated: where the plugin is absent, say so inline, state what it would have owned, and record the fallback in Routed-to.

Executing a finding belongs to overengineering:realign, which presents this lane's rows and offers no rung for them until a ladder for these layers exists. Name it as the next step and never start it unasked.

The report

${CLAUDE_PLUGIN_ROOT}/skills/audit/context/report-template.md owns the output shape: the findings artifact as the single source of truth, an inline summary always. Three additions for this lane. A widened target puts the widening in the first line, before anything else. A routed target is reported inline with the layer that claimed it and no row at all, so the operator can tell a route from a skip. The template's "always" on writing the artifact carries this lane's two exceptions, both in the read-only contract above: no branch identity, and a run that filed no row, however it got there. Because this lane cannot know at line-one time which it will be, the read-only opening line names the resolved home rather than asserting a write, per the template's item 1, and the closing summary states whether anything was persisted.

Gotchas

• A single grep is not a search. A multi-word phrase wrapped across a line break defeats a single-line match, and the file that would have refuted the finding is the one that wraps it. Vary the wrapping, the hyphenation, the casing, and the obvious synonyms before "not found" becomes a finding, and record which forms were tried.
• Relocated coverage is not absent coverage. A guard that never fires because the concern it guards is now handled earlier is not an unguarded concern. Find where the concern went before calling it gone; the finding is then about the duplicate, argued on carry cost.
• Silence about a document is usually unavailable, not empty. Most consumers record nothing about which documents get read, so the honest reading of no usage data is that tier 1 does not exist here (§2). Collapsing the two manufactures confidence.
• A class match is a marker, never a verdict. "It is a decision record" says which rules apply, not whether this one earns its keep.
• The operator's framing is not evidence. A target arriving with its conclusion attached ("this is duplication", "retire it") is a hypothesis to test, and agreeing with it is not a finding.