Guru Reconcile Task Base

---
name: guru-reconcile-task-base
description: Reconcile one exact active-task and evolved-base pair through a semantic impact review, controlled integration evidence, and six typed exits.
---

# Reconcile Task Base

Invoke this Skill only after its package-local pair guard returns `new_pair` at
an eligible stable workflow boundary. Read `references/contract.md`, then bind
the exact caller profile, base pair, task authority, planning, and candidate
evidence before forming the AI judgment.

Use the exact caller profile and pair supplied by the boundary. Treat the base
pair as an integration clock only: its advance does not by itself invalidate
approved planning or an already selected workflow stage. Build only a
temporary candidate through the declared deterministic command. After the AI
has separately judged authority, task-content, and integration impact, record
and check that result, then return exactly one declared typed exit. Pre-review
profiles with unchanged authority/task content and a compatible candidate
return `reconciled` with the original `resume_target`, including `post_plan`.

For `post_branch_review`, `post_publication`, and
`finalizer_base_mismatch`, return `review_continuity_required` when the prior
full review remains valid but the compatible candidate needs a new
reviewed-content identity. After the semantic gate, display the exact task
branch, prior task HEAD, selected new base, candidate tree, local merge commit
message, and zero remote effects. Invoke `execute-base-reconciliation` only
after the user confirms that exact action in the current dialogue. The
executor creates one expected-head-bound local merge commit and never pushes;
the confirmation is never passed to or persisted by a script. Record the
checked executor result, then hand Branch Review the prior full-review commit
separately from the current reconciled task HEAD.

Before a base-delta observation can become an impact, finding, validation
obligation, scope question, implementation route, or stale-plan route, form its
candidate-only set and invoke `guru-qualify-normal-scenario` with
`base_impact_candidate_set`. Assign no severity or route before `classified`.
Rejected candidates do not trigger clarification or task work. Mechanism
revision returns here for remove/replace and fresh qualification; blocked
stops. No qualification result is persisted in the base checkpoint.

Before a proposed integration mechanism affects the impact judgment,
validation obligation, or route, invoke `guru-qualify-solution-mechanism` with
`base_impact_candidate_set`. A `mechanism_revision_required` result removes or
replaces only that mechanism and returns here for fresh qualification; it is
not a scope-confirmation path.

Mapped exits and stale re-entry are automatic. Ask the user only when the
semantic result is `scope_confirmation_required`; never persist that
confirmation. Missing, stale, ambiguous, non-ancestor, or unmapped evidence
fails closed.

For the conflict-resolved path, run the resolved executor and then
`project-resolved-full-review`. That deterministic package command emits the
complete existing `guru-review-branch:branch_review` input directly; it is not
an additional Reconcile typed exit and never enters Task Commit or bounded
base continuity.