Task — engineering-spec@1

task-author manifests default to untracked .workflow session state

doneTASK-IMP-090
module improvement · class product · priority p2 · created 2026-07-16 · shipped 2026-07-17
depends on none · blocks none

TASK-IMP-090: task-author manifests default to untracked .workflow session state

Summary

Author-run manifests are session state, but the skill defaults them into the tracked task tree and the .workflow gitignore seed only covers ship manifests - so this run's three batch manifests landed tracked. Per the recorded IMP-11 decision: the skill default moves to docs/tasks/.workflow/, the gitignore seed grows *.manifest.json, the three tracked manifests leave the index, and a tracked batch summary in _audits/ becomes the durable approval record.

Problem

Session caches in tracked space churn history and get mistaken for records; the actual record the operator needs (what was planned, who approved, what landed) belongs in _audits/ like the sachviet run's batch summary.

Proposed Solution

Four small moves: (1) SKILL.md CONTRACT_ECHO default becomes docs/tasks/.workflow/task-author.<slug>.manifest.json; (2) install.sh's .workflow/.gitignore seed writes both *.ship.json and *.manifest.json, and appends the manifest pattern once to an existing seed that lacks it; (3) git rm --cached the three batch manifests in this repo and update its .workflow/.gitignore; (4) write docs/tasks/_audits/IMPROVEMENT-BATCHES-2026-07-16.md summarizing batches 1-3 (members, verdicts, evidence commits) as the tracked approval record.

Alternatives Considered

Success Metrics

Scope

In scope: the skill default line, the seed patterns plus append-once migration, this repo's index cleanup, the _audits summary.

Out of scope / Non-Goals

Dependencies

AI Authorship Disclosure

1. Description (normative)

2. Acceptance criteria

3. Edge cases

Audit

TASK-IMP-090-untracked-author-manifests spec audit - audit_rubric@2.0 (machine floor + judgment)

Machine floor: task-lint clean on first pass (FM/SEC/COND/TRACE-structural all green). Judgment families: metrics carry baseline/target/deadline grounded in recorded run or incident evidence; alternatives distinct with real rejection reasons; scope subsections present; sourced untrusted evidence only where quoted; ops-verified ACs carry explicit rationales; COND-004 bullets truthful.

ISSUE ISS-001 (QA-004, wontfix-info): several ACs are ops-verified (prose contracts, repo-state chores, single documents) with recorded-evidence rationales judged sufficient - the same pattern accepted for TASK-IMP-086/087 and consistent with TRACE-002's manual allowance.

SUMMARY verdict: pass issues_open: 0 issues_human: 0 next_action: ship

§gate-log

Populated during implementation (ship-tasks testing phase).