Agent-Memory Workflow Migration Manifest (Bucket 2)

Mission workflow-mechanics-self-doc-01M02SF1 is the "Bucket 2" follow-up to self-documenting-repo-01M0287X (Bucket 1, see agent-memory-migration-manifest.md). Where Bucket 1 migrated the operator's gate-remedy / recovery-runbook / env-friction memory entries (G1–G6), Bucket 2 audits the remaining ~49 workflow, CI, git, and status/mission/sync mechanics entries — product mechanics an agent still has to learn the hard way when no canonical doc says them for itself.

The audit that produced the disposition for each entry (work/bucket2-workflow-memory-audit.md) is gitignored and does not exist in this worktree or in CI, exactly like Bucket 1's working file — so this manifest is the committed authority for the migration, not a pointer to one. The companion test, tests/docs/test_workflow_migration_manifest_complete.py, parses the three cluster tables below at runtime and enforces that every row carries a recognised token and every path-bearing token resolves to a real file — it is the enforcement half of this manifest, not just a report of it.

Out of scope (C-004, same as Bucket 1): deleting the resolved entries from the operator's live MEMORY.md file is a manual, per-operator checklist tracked on #3448. This manifest produces the resolution map; the operator applies it at their own pace.

Resolution vocabulary

Each row below carries exactly one resolution token:

Token	Meaning
`home:`	The lesson now lives at this repo path because a WP in this mission created the file or added the section that says it for itself.
`already-home:`	The lesson was already covered by an existing repo path before this mission touched anything — the memory entry is a deletable echo.
`learned-fact:`	Too narrow or environment-specific for a doc; captured as a git-tracked, team-shared note under `.kittify/memory/`.
`keep-private`	Operator-working-style calibration or single-agent-harness behaviour; deliberately not codified into a shared repo surface.
`charter-candidate`	The lesson proposes a new governance rule; routing to the charter is operator-gated and not assumed by this manifest.

Cluster A — Landing / git (17)

Memory entry	Resolution
`feedback_fork_pr_landing_push`	already-home: `GIT_WORKTREE_PR_WORKFLOW.md` — "Fork PR pushes go over SSH" is already codified harness-neutral; the replacement-PR half is in `pr-landing.md` §9.
`feedback_force_with_lease_rejection`	already-home: `GIT_WORKTREE_PR_WORKFLOW.md` — "(stale info) = origin moved; fetch + cherry-pick, never plain `--force`" is verbatim there.
`feedback_rebase_check_before_landing_push`	already-home: `pr-landing.md` §3 (rebase-first) + §9 (fetch-before-push).
`feedback_reuse_clone_per_mission_branch`	keep-private — "prefer one clone + branch-per-mission" is an operator clone-vs-worktree preference; the safety half is the separately-resolved `feedback_one_mission_per_checkout`.
`feedback_no_branch_switch_during_bg_tests`	already-home: `GIT_WORKTREE_PR_WORKFLOW.md` — "don't move a worktree's HEAD while a background job reads it" is present with rationale.
`feedback_one_mission_per_checkout`	home: `git-worktrees.md` — "Cross-Mission Concurrency Is a Different Question" section (WP05): two missions in one checkout race the shared git HEAD/index.
`feedback_no_git_stash_in_lane_worktrees`	already-home: `GIT_WORKTREE_PR_WORKFLOW.md` — the shared `refs/stash` stack across worktrees is documented, plus `known-friction-points.md`.
`feedback_isolate_pr_review_agents`	already-home: `pr-landing.md` §2 (one isolated worktree per PR) + §8 (read-only landing worktree).
`feedback_isolation_review_needs_pushed_commit`	keep-private — Claude-Code-harness `isolation:"worktree"` behaviour (seeds from a pushed commit), not a Spec Kitty product mechanic.
`feedback_isolation_worktree_seeds_from_primary_repo`	keep-private — same harness-specific worktree-seeding detail; kept out of the harness-neutral toolguide.
`feedback_claim_pr_before_working`	already-home: `pr-landing.md` §1 "Claim before touching" — verbatim.
`feedback_pr_landing_plus_lightweight_squad`	already-home: `pr-landing.md` §8 "Adversarial squad" — the 2-lens floor, fold-by-default, and immediate-file discipline are all present.
`feedback_landing_pass_lessons_0704`	already-home: `pr-landing.md` §9 (lease via `rev-parse`) and §3 (changelog symlink); the "Closes-partial-issue" half is the separately-resolved `feedback_pr_closing_keyword_parsing`.
`feedback_small_fix_pr_with_charter_files_is_stale_stack`	home: `pr-landing.md` — "Stale-stack diagnostic: two-dot vs three-dot diff" subsection (WP04): a small-fix PR carrying charter files under a two-dot diff but not a three-dot one is smuggled governance.
`feedback_pr_closing_keyword_parsing`	home: `GITHUB_TRACKER.md` — "Closing keywords — one issue per keyword" section (WP08): `Closes #A,#B` only auto-closes `#A`; repeat the keyword per issue.
`reference_history_compression_by_path_bucket`	home: `compress-mission-history.md` (WP04, new how-to): the path-bucket `git commit-tree` snapshot-chain recipe + tree-parity proof.
`feedback_post_merge_history_compression`	home: `compress-mission-history.md` — same how-to; the "never `rebase -i`" reasoning and phase-boundary technique are the other half of the same recipe.

Cluster B — CI (14)

Memory entry	Resolution
`feedback_collect_universe_once_reuse`	keep-private — corrected this run: the audit's working draft proposed `learned-fact:`, but WP09 did not create a matching `.kittify/memory/` note, so that path-check would fail. The anti-pattern is unique to `_gate_coverage.collect_universe` (already fixed in code) and too narrow for a doc; kept private rather than routed to a nonexistent file.
`feedback_precision_runs_not_full_suite_local`	keep-private — the general principle (CI owns the full sweep) is already home in `CLAUDE.md`/`testing-parallel.md`; "only targeted runs locally" is operator-environment calibration.
`feedback_no_full_arch_suite_locally`	keep-private — environment-specific to the operator's machine, and in tension with `pr-landing.md`'s "run `tests/architectural/` locally on the rebased tip"; kept private rather than codifying a machine-specific limit.
`reference_ci_label_skip_guard`	home: `known-friction-points.md` (WP03): the `pr:deferred` / `pr:skip-ci` job-skip guard, cited to `ci-quality.yml`.
`reference_red_main_ci_release_policy`	already-home: `red-main-and-release-readiness.md`, plus ADR `2026-07-17-1` and charter Standing Order #9 — fully codified across multiple surfaces already.
`feedback_classify_reds_against_true_mission_base`	home: `pr-landing.md` — "Multi-WP lanes: classify against the true base, not the lane tip" subsection (WP04): `git merge-base <mission-branch> upstream/main`, not the lane tip.
`feedback_path_filter_unskips_suites_on_new_path_push`	already-home: `pr-landing.md` — "shard path-filters mask pre-existing failures… your own folds trigger this too" is documented verbatim.
`feedback_module_move_into_critical_path_coverage`	home: `coverage-signals.md` (WP03) — "Remedy: `git mv` into a critical-path dir needs `fast`-marked coverage" section.
`feedback_green_pr_regardless_of_red_origin`	already-home: `pr-landing.md` §4 — "fold the fix in by default… keep main green even for a red the PR did not cause."
`feedback_ci_fix_landing_discipline`	already-home: `pr-landing.md` §4, plus `review-gates.md`'s Typer/click version-skew section and `known-friction-points.md`'s stale-install entry.
`feedback_verify_lint_inputs_tracked`	home: `known-friction-points.md` (WP03) — `charter lint`'s project-DRG input (`.kittify/doctrine/graph.yaml`) is gitignore-adjacent; confirm it is tracked and in-diff before trusting a `charter lint` result.
`feedback_no_backticks_in_shell_double_quotes`	already-home: `EFFICIENT_LOCAL_TOOLING.md` — "Shell quoting" section: backticks inside double quotes run as command substitution; use `--body-file`/heredoc.
`feedback_dispatched_impl_foreground_tests`	already-home: `EFFICIENT_LOCAL_TOOLING.md` — "Gate execution mode" section: run gates in the foreground; backgrounding stalls the handoff.
`feedback_no_recursionerror_is_not_no_cycle`	learned-fact: `no-recursionerror-is-not-no-cycle.md` — narrow verification discipline: absence of a crash does not prove absence of a cycle; trace the call graph.

Cluster C — Status / mission / sync (18)

Memory entry	Resolution
`reference_issue_matrix_approval_gate_mechanics`	home: `review-gates.md` (WP01) — "Issue-matrix discovery and `issue-verdict --actor`" section covers the genuinely-absent discovery-surface and `--actor`-required half; the verdict vocabulary/schema/`.json`-canonical rule was already-home at `ERROR_CODES.md` and `spec-kitty-mission-review/SKILL.md` (Gate 4) before this mission and is cited from `review-gates.md` rather than restated.
`project_in_mission_issue_matrix_verdict`	already-home: `spec-kitty-mission-review/SKILL.md` — the non-terminal `in-mission` verdict (accepted at `approved`, rejected on `done`) is already documented there.
`reference_flat_missions_still_require_lanes`	home: `execution-lanes.md` (WP05) — the `MissingLanesError` contract; the stale `AGENTS.md`/`CLAUDE.md` line that claimed a retired `lanes.json`-absent legacy fallback was fixed in the same pass (WP07).
`feedback_lane_move_task_sync_hang`	already-home: `known-friction-points.md` — "`move-task` can hang on sync-daemon fan-out" entry, plus `review-gates.md`'s pre-review regression-gate section.
`reference_merge_review_artifact_invariant`	home: `review-gates.md` (WP01) — "Review-cycle artifacts and the merge gate" section: `terminal_wp_latest_review_artifact_must_not_be_rejected`, its shared implementation across `merge`/`merge --dry-run`/`review` Gate 1, and the schema. Corrected this run: the original memory entry's framing (a frontmatter-based mechanism) describes retired behaviour — see "Corrections surfaced this run" below.
`reference_sync_identity_form_split`	learned-fact: `sync-identity-form-split.md` — corrected this run: the original memory cited issue #883 as the tracking reference; that citation does not resolve to this defect and has been dropped rather than carried forward.
`reference_sync_drain_gate_chain`	home: `sync-drain.md` (WP02, new runbook) — the 3-gate drain order (`saas_disabled` → `missing_auth` → `missing_team`) and the `sync doctor` false-green trap.
`feedback_seam_classification_two_axes_and_gate_vacuity`	home: `artifact-placement-seam.md` (WP05) — "Two-Axis Resolver-Site Classification" (Axis A raise-or-degrade, Axis B anchor-root); the gate-vacuity half ("prove the gate reads its authority at runtime—an alias/re-export defeats it") is the sibling addition in `architectural-gate-non-vacuity.tactic.yaml` (WP08).
`reference_review_cycle_artifact_frontmatter_trap`	home: `review-gates.md` — consolidated into the same "Review-cycle artifacts and the merge gate" section: `move-task --review-feedback-file` provenance guard, cycle numbering under lock.
`feedback_review_artifact_no_hand_author`	home: `review-gates.md` — "Never instruct an agent to hand-write an `approved` review-cycle artifact to unblock a gate" is stated directly in the merge-gate section (WP01); the doctrine-layer rationale (two-party review integrity) is the sibling enrichment "Never manufacture or self-certify an approved artifact" in `reviewer-implementer-role-separation.tactic.yaml` (WP08). The charter-routing recommendation the original audit flagged for this entry is superseded by that tactic shipping — no open `charter-candidate` remains for it.
`project_lane_bare_python_imports_primary`	already-home: `EFFICIENT_LOCAL_TOOLING.md` — "Python environment isolation (uv)" section, plus `known-friction-points.md`: a bare `python`/`pytest` in a lane or clone imports the PRIMARY `src`; always `uv run`.
`reference_rejected_then_fixed_approval_override`	home: `review-gates.md` — "The `--skip-review-artifact-check` escape hatch" subsection: the arbiter-override path, mandatory `--note`, and the `ReviewOverride.complete` gating condition.
`feedback_read_write_partition_symmetry`	home: `artifact-placement-seam.md` (WP05) — "Partition-Move Audit Checklist": when a partition classification moves, grep every reader, classify both axes, and prove it with an e2e test rather than a unit read alone.
`feedback_seam_bypass_stale_comment`	home: `canonical-source-unification.tactic.yaml` (WP08) — "Seam-bypass via stale comment" failure mode: a "can't reuse the seam" comment is a red flag to verify, not a fact to trust.
`feedback_lane_base_guard_vs_moving_upstream`	learned-fact: `lane-base-vs-moving-upstream.md` — narrow dispatch heuristic: guard a lane base against a fixed rebase-target SHA, not a moving `is-ancestor upstream/main` check.
`feedback_no_legacy_resolver_paths`	already-home: `2026-07-01-1-no-legacy-compat-branches-in-resolvers.md` — the exact rule (require canonical + migrate; forbidden `if <field> is None: <legacy fallback>`) is that ADR's whole subject.
`feedback_gate_unmask_cannot_self_validate`	already-home: `quality-and-tech-debt-standing-orders.md` — "A gate-unmask cannot self-validate… pair every un-mask with a pre-merge full-gate dry-run; never ship a mission-diff-scoped assertion to the main branch."
`feedback_execution_state_impl_discipline`	already-home: `030-test-and-typecheck-quality-gate.directive.yaml`, plus directives `025-boy-scout-rule`, `034-test-first-development`, and `041-tests-as-scaffold-not-friction` — the tagged-and-reduced-tests, ruff/mypy-clean, and fix-don't-rationalize pointers echo existing directives.

Corrections surfaced this run

Running this audit against the live, shipped state of the WP01–WP09 work (rather than trusting the audit's own draft text) surfaced four cases where the working draft was itself stale or imprecise — the mission's thesis proven in miniature, same as Bucket 1's four corrections:

Review-cycle / merge-gate mechanism is event-sourced, not frontmatter-based. The audit's working draft framed reference_merge_review_artifact_invariant around parsing on-disk review-cycle-N.md frontmatter. WP01 found the shipped gate (terminal_wp_latest_review_artifact_must_not_be_rejected) reads the reduced status snapshot's event-sourced review/review_result slots and never parses on-disk frontmatter; a hand-edited review-cycle-N.md has zero effect on the gate. review-gates.md documents the current mechanism, not the retired one the memory described.
Sync-drain gate order corrected: the drain is saas_disabled → missing_auth → missing_team, and spec-kitty sync migrate is not one of the three gates — it is a separate, mostly-obsolete legacy-queue command that some older notes mislabelled as "gate 2". sync-drain.md states this explicitly rather than repeating the stale framing.
reference_sync_identity_form_split's #883 citation was a mis-citation. The issue number did not resolve to the identity-form split it was attached to in the working draft; rather than carry a dangling reference forward, the .kittify/memory/ note drops it.
feedback_collect_universe_once_reuse reclassified from learned-fact: to keep-private. The audit's working draft proposed a .kittify/memory/ note for this entry, but WP09 (the learned-facts seeding WP) did not create one — the underlying anti-pattern was already fixed in code and judged too narrow for a shared note. Routing it to learned-fact: here would point the manifest's own path-check at a file that does not exist; keep-private reflects what actually shipped.