TS — TFW_20260830-114238_ASSISTED15: Prompt/File Assisted 1.5 Product and Maintenance Bridge¶

Date: 2026-08-30 Author: saubakirov via Codex Coordinator Status: ✅ APPROVED — owner amendment A1 and bounded autonomous mandate, 2026-08-30 Parent HL: HL-TFW_20260830-114238_ASSISTED15 Contract baseline: re-freeze commit with reserved freeze scope after amendment A1 Research: Iteration 1 RES · Iteration 2 RES, with executable-runtime conclusions superseded by A1

1. Objective¶

Deliver the single amended phase as a complete public Assisted 1.5 under editions/: a standalone neutral Russian starter whose TFW mechanics are prompts, Markdown skills, static contracts and ordinary agent file operations. Remove the executable identity/registration and maintenance/update/self-test runtime introduced by the earlier implementation. Preserve the useful templates, truthful Assisted changelog, static release authority, asymmetric forward/update promotion route, downstream state and read-only Innoforce evidence. шаблоны/build_a4.py may remain only as an artifact-template builder.

2. Scope¶

In Scope¶

Keep the five complete Assisted skills and metadata, but make tfw-identity a Markdown-only prompt/file procedure and /tfw-update an agent-led compare, gate, file-update and promotion procedure.
Delete editions/02-assisted/.agents/skills/tfw-identity/scripts/tfw_identity.py and editions/maintenance/assisted_maintenance.py; add no replacement workflow executable.
Remove every product command, claim, metadata dependency and shipped test contract that requires the deleted runtime.
Retain and regenerate the static release manifest and maintenance policy as human/agent-readable authorities rather than executable-engine inputs.
Preserve the complete neutral Russian note, work-plan, A4 and presentation templates; keep build_a4.py strictly inside artifact generation.
Preserve the exact retired-hook deletion, public Assisted VERSION/CHANGELOG, edition guidance and installed-state protection rules.
Regenerate task-local RF/EV/evidence for the amended no-code product, including prompt/file scenario evidence and both maintenance directions.
Make scoped local commits; after terminal RF, stop at a clean boundary for the Coordinator's one-time history compaction before final independent review.

Out of Scope¶

Any product edit outside editions/, including .tfw/, root guides, KNOWLEDGE.md, TECH_DEBT.md, editions/01-light/, or unrelated task folders.
Any executable TFW registration, identity, lifecycle, maintenance, update, locking, ACL, synchronization or product self-test subsystem, regardless of language.
Expanding шаблоны/build_a4.py beyond A4 artifact generation or using it to verify/operate TFW mechanics.
Any write, rename, cleanup, generated artifact or metadata change under H:\Shared drives\IT\Innoforce AI-First Knowledge\innoforce_starter_v1.5 or adjacent preserved versions.
Copying Innoforce knowledge, people, facts, products, branding, paths, identifiers, release history or unique operational context into the public package or public evidence.
Automatic mutation of the current mixed field tree, a symmetric mirror, source authentication/signing, a distributed-transaction claim, English localization, push, tag creation/deletion/rewrite or remote publication.
Deleting or rewriting the external local tag v2.0.0-dirty.5; final evidence records it as concurrent external state.
Reopening another frozen claim, accepting a path/LOC overrun, or modifying unrelated TFW-55/TFW-60 work without a new owner ruling.

3. Principles Check¶

#	Principle (from amended HL §7)	Enforced by	Gate
P1	Field practice is evidence, not product authority	AC-2, AC-6, AC-8	Human behavior is retained; field runtime and company context are absent
P2	Templates are product behavior	AC-7	Complete worked templates render offline and are visually inspected
P3	Neutral core, explicit overlay	AC-4–AC-6	Static authorities and skill instructions distinguish public, customizable and downstream paths
P4	One static authority per file class	AC-4, AC-5	Every maintained path has one declared owner and one documented rule
P5	Promotion, not blind synchronization	AC-5, AC-6, AC-10	Forward work is gated; reverse work is candidate-only and independently reviewed
P6	Standalone by default	AC-1, AC-2	A copied starter works without Full, Light, Innoforce, registry or hidden service
P7	Human authority and honest automation	AC-2, AC-3, AC-5	Ambiguity asks/stops; optional orchestration never replaces acceptance gates
P8	Stable trace over moving state	AC-3, AC-9, AC-10	Same role sessions and durable task traces survive amendment and compaction
P9	Prompt and files over workflow runtime	AC-1, AC-2, AC-4, AC-8	Product code inventory permits only the artifact builder
P10	Functional completeness over byte identity or executable complexity	AC-1, AC-7–AC-10	Useful behavior and outputs are evidenced without copying field bytes or shipping a mechanics engine

4. Affected Files¶

File	Action	Description
`editions/README.md`	MODIFY	Accurate Assisted 1.5 choice and no-code capability boundary
`editions/02-assisted/{AGENTS.md,README.md,PROJECT.md,MIGRATION.md}`	MODIFY	Neutral onboarding, project-unknown state, prompt/file mechanics and migration contract
`editions/02-assisted/{people/README.md,knowledge/INDEX.md}`	MODIFY	Profile/file procedure and empty knowledge navigation with no shipped participant state
`editions/02-assisted/{VERSION,CHANGELOG.md}`	CREATE	Exact public version authority and truthful public-only history
`editions/02-assisted/.codex/{hooks.json,hooks/tfw-hook.ps1,hooks/tfw-hook.sh}`	DELETE	Exact retired stock lifecycle-hook payload
`editions/02-assisted/.agents/skills/{tfw-plan,tfw-handoff,tfw-review,tfw-update,tfw-identity}/{SKILL.md,agents/openai.yaml}`	CREATE	Five complete prompt/Markdown role contracts and metadata
`editions/02-assisted/.agents/skills/tfw-identity/scripts/tfw_identity.py`	DELETE FROM CURRENT TASK RESULT	Remove executable registration/identity runtime; final path is absent and therefore not a baseline product-diff path
`editions/ASSISTED_MAINTENANCE.md`	CREATE	Static authority classes and agent-led compare/update/promotion route
`editions/maintenance/{release-manifest.json,maintenance-policy.json}`	CREATE	Static release inventory and maintenance policy
`editions/maintenance/assisted_maintenance.py`	DELETE FROM CURRENT TASK RESULT	Remove executable maintenance/update/self-test runtime; final path is absent and therefore not a baseline product-diff path
`editions/02-assisted/шаблоны/{заметка.md,план_работы.md,документ_A4.md,презентация.html,build_a4.py,theme.css}`	CREATE	Complete neutral Russian practical templates and artifact builder
`editions/02-assisted/шаблоны/overlay/theme.css`	CREATE	Stock-customizable six-property artifact theme overlay
`editions/02-assisted/шаблоны/assets/tfw-mark.svg`	CREATE	Neutral shape-only artifact mark

Final product budget from f3eb986: at most 33 paths—23 new, seven modified and three deleted—and at most 2,600 changed product lines. The two removed runtime files are current-task deletions but disappear from the final baseline diff. Any 34th final path, 24th new file or line over 2,600 is a hard stop for a new owner amendment.

5. Acceptance Criteria¶

AC-1: Release boundary, standalone package and no-code product¶

The final edition is a complete uninitialized public Assisted 1.5, and TFW mechanics ship no executable runtime.

[ ] Product diffs attributed to this task are confined to editions/; no Assisted task commit changes .tfw/, root guides or editions/01-light/.
[ ] VERSION contains exactly 1.5\n; CHANGELOG.md truthfully records public Assisted 1.0 and 1.5 without importing Innoforce history.
[ ] A clean copy starts without profile, binding, project_id, work, task trace, organization record, registry or hidden Full/Light/Innoforce dependency.
[ ] tfw_identity.py and assisted_maintenance.py are absent. No product .py, .ps1, .sh, .js, .ts, .bat or .cmd implements TFW mechanics; шаблоны/build_a4.py is the sole permitted Python product file and only builds artifacts.
[ ] Final product diff is no more than 33 paths, 23 new files and 2,600 changed lines, with the expected three hook deletions.

Gate: Exact task-attributed Git census from f3eb986, product executable inventory, private-marker scan and clean-copy inspection. Evidence: Retain the path/action/LOC census, executable semantic inventory, version bytes and clean-copy file tree in EV.

AC-2: Prompt/file identity and profile safety [depends: AC-1]¶

Identity is a Markdown procedure performed by the agent with ordinary reads, questions and file creation; it is not registration software.

[ ] tfw-identity/SKILL.md enumerates valid people/*.md profiles, uses the explicit one-profile rule only when exactly one valid profile exists, and otherwise asks the human rather than guessing.
[ ] Participant, organization role, project role, task owner and AI role are separate fields; changing a session participant or role never changes an existing task owner silently.
[ ] New profile creation requires explicit name, surname and roles; derives the documented surname identifier; checks the exact target immediately before creation; never overwrites a collision; rereads the created file.
[ ] Zero, one and multiple profile scenarios; Cyrillic/Latin surname; missing surname; collision; invalid profile; explicit fixed selection; and autonomous no-human-role behavior are complete in the written contract.
[ ] No machine binding, local registry, device namespace, ACL probe, lock file, temporary identity store or OS-security promise remains in product text, metadata or paths.

Gate: Independent semantic walkthrough of the Markdown skill and scenario table against isolated synthetic people/ trees; before/after manifests prove only the explicitly approved profile file changes. Evidence: Record each prompt, available profiles, user-supplied choice, expected file action and observed result; do not execute a shipped helper.

AC-3: Complete lifecycle and reusable role boundary [depends: AC-1]¶

The five skills implement plan → handoff → independent review → human acceptance with a complete manual path and optional capability-gated orchestration.

[ ] All five SKILL.md and metadata pairs are complete, consistent and reference only paths present in the copied starter.
[ ] Stable task paths, Coordinator/Executor/Reviewer role locks, human acceptance, knowledge decision and manual/autonomous choice are explicit.
[ ] Missing capability, ambiguous handle or unconfirmed interruption stops or switches to manual-complete without spawning an automatic duplicate.
[ ] Correction reuses the same Executor; re-review reuses the same independent Reviewer and reruns the full amended contract; child reports go only to the Coordinator.
[ ] No lifecycle hook, startup registration, hidden product test command or automatic enforcement claim remains.

Gate: Contract/link scan plus a written deterministic role scenario matrix for complete, partial, lost-handle, no-interrupt, overlap, manual fallback and full re-review. Evidence: Use this task's actual one-Coordinator/same-Executor/same-Reviewer trace and a manual-complete tabletop; no product runtime is required.

AC-4: Static release and maintenance authority [depends: AC-1]¶

The manifest and policy are static, deterministic, readable authorities; they are not an executable maintenance protocol.

[ ] release-manifest.json self-excludes, includes maintenance-policy.json, and lists every remaining public payload file exactly once with relative path, size and SHA-256.
[ ] maintenance-policy.json declares release version, accepted public baseline, the three exact retired hook hashes, public/customizable/downstream authority classes and target-only preservation.
[ ] JSON parses with no duplicate/colliding/traversal/absolute paths; paths are normalized text and resolve within editions/ during task-local verification.
[ ] /tfw-update and ASSISTED_MAINTENANCE.md describe how the agent reads these records, compares actual files, stops on ambiguity/drift and presents one gated change plan.
[ ] No product command parses, validates, journals or applies the records automatically.

Gate: Two independent task-local regenerations of the static row set produce identical rows and hashes; semantic review confirms the policy has one authority per path class. Evidence: Retain the two generated row sets/digests and the reviewer-readable authority table; task-local tools may verify but are not shipped in editions/.

AC-5: Agent-led forward update and installed-state preservation [depends: AC-4]¶

Forward update is a reviewed sequence of ordinary file operations, not an engine or transaction claim.

[ ] The update skill requires source manifest/policy, complete target inventory and exact baseline comparison before proposing any write.
[ ] The plan classifies every path as public, customizable, downstream-only, retired stock or unresolved; unresolved/drifted state stops before change.
[ ] One explicit gate identifies the exact files to create/replace/delete; after approval the agent rechecks each precondition, uses ordinary file tools and captures a post-change manifest.
[ ] work/, knowledge, people/profiles, project identity, personalization, unrelated .codex, unknown target-only paths and modified templates remain byte-identical unless one explicit migration rule covers them.
[ ] A partial ordinary-file update is reported as partial and returned to review; documentation never calls the sequence atomic or distributed-transaction safe.

Gate: A clean synthetic downstream fixture and a drifted/customized fixture are executed by the same Executor using only the documented skill/file procedure. Evidence: Before/plan/approval/after tables show every path, preserved bytes, stopped drift and zero unexplained change; no assisted_maintenance.py invocation exists.

AC-6: Reverse promotion privacy and field immutability [depends: AC-4]¶

Reverse work produces a reviewed generic candidate, never a mirror or direct public mutation.

[ ] The update skill compares downstream/public inventories, classifies generic versus organization-specific differences and prepares a candidate outside the public core.
[ ] Candidate text contains no private path, hash, person, organization, brand, project, release-history or unique operational context.
[ ] Independent semantic review is mandatory before any candidate can become a later public product change; this task does not auto-apply it.
[ ] The real field lineage is read-only comparison evidence and its 29-row tree is identical before and after all task work under both documented digest orders.

Gate: Synthetic generic/private reverse cases plus a non-mutating comparison of the pinned field source. Evidence: Retain candidate/privacy verdicts and field pre/post row equality; never retain private source contents in public product files.

AC-7: Useful neutral offline templates [depends: AC-1]¶

The practical template set remains complete, generic, Russian-authoritative and readable offline.

[ ] Note, work plan, A4 source and presentation are complete worked generic examples rather than placeholders.
[ ] build_a4.py accepts only artifact inputs/outputs/theme/title, resolves local resources within the template tree and contains no TFW registration, identity, update, maintenance or test behavior.
[ ] Theme overlay and SVG mark retain their closed passive interface; external/network/active resources reject.
[ ] Stock/custom A4 and presentation render with network blocked; every produced page/capture is visually inspected for readability, correct file type, no local URL/header and no private marker.

Gate: Static attack fixtures, two artifact-builder runs, blocked-network render, page/signature checks and visual inspection. Evidence: Save current rendered outputs under task evidence and record page-by-page inspection; the builder is assessed solely as an artifact tool.

AC-8: Cross-file agreement, neutrality and removed-runtime closure [depends: AC-2, AC-3, AC-4, AC-7]¶

Every public claim agrees on the amended prompt/file product and contains no stale runtime or downstream residue.

[ ] Edition/root docs, skills, migration, people guide, templates, changelog, manifest and policy agree on exact paths, version, prompt identity and agent-led update semantics.
[ ] No command or link names either removed Python runtime; no text requires a local registry, OS lock, ACL proof, executable verifier or automatic maintenance operation.
[ ] No Innoforce/company/person/product/leadership/value/location/path/brand/logo/project default, private hash or field-only release heading appears in product text, metadata, assets or generated output.
[ ] Public 1.0 baseline and 1.5 amendment history are truthful; no product text invents a tag, SemVer, English mirror or private provenance.

Gate: Full text/link/code inventory, removed-path reference scan, binary/SVG inspection, synthetic private-token corpus and independent semantic review. Evidence: EV names every removed runtime reference and its corrected public statement; zero hits are resolvable, not asserted from memory.

AC-9: Proportionate no-code evidence package [depends: AC-2–AC-8]¶

Task-local evidence proves the amended product without shipping a product verification engine.

[ ] Evidence covers the exact product census, no-code inventory, static manifest regeneration, prompt identity scenarios, lifecycle/role scenarios, forward preservation, reverse privacy, template renders, cross-file agreement and source immutability.
[ ] Every negative scenario proves zero unauthorized product/fixture mutation through before/after manifests or documented no-write observation.
[ ] Repeated static inventory and render checks produce the same public hashes; nondeterministic environment facts are recorded rather than hidden.
[ ] RF/EV explicitly supersede earlier runtime evidence and prior REVISE conclusions; no stale FINAL=PASS, ACL, lock or no-tag-containment claim is reused as current proof.
[ ] Task-local evidence scripts, if retained, live only under this task, are not referenced by the product and test files rather than operate user workflow state.

Gate: Complete EV row audit, task schema validation, task-attributed Git audit and independent attachment resolution. Evidence: Every VERIFIED row cites a current attachment; previous evidence remains historical but is not counted as amended acceptance.

AC-10: Both directions, truthful publication state and final review [depends: AC-5, AC-6, AC-9]¶

The amended result closes both maintenance directions and the task lifecycle without rewriting external state.

[ ] Agent-led clean public→downstream fixture reaches Assisted 1.5 with protected state unchanged; downstream-generic→public produces only a reviewed candidate.
[ ] No Assisted task operation performs push, tag creation/deletion/rewrite or remote publication. Evidence truthfully records that concurrent external local tag v2.0.0-dirty.5 contains Assisted ancestors and that no remote-tracking ref contains the terminal result at capture time.
[ ] Same role lineage is preserved through amendment: one Phase Coordinator, same Executor, same independent Reviewer, with a full final rerun against the re-frozen HL and revised TS.
[ ] After terminal RF/evidence and before final review, the Coordinator performs history compaction exactly once on a recoverable local backup branch (not a tag), preserves all unrelated TFW-55/TFW-60 commits and dirty files, and proves pre/post tree equality.
[ ] Final compact history contains logical owner-config, task plan/research/final TS, final Assisted product, evidence/RF with preserved review traces, and final approval/closure commits; obsolete task checkpoint/fixup commits are absent.

Gate: Direction evidence, source digest, Git operation-attribution audit, one-time compaction attestation, exact tree comparison and full independent /tfw-review. Evidence: Retain compact commit mapping, backup branch name, old/new tree IDs, unrelated commit preservation audit, current tag/remote containment facts and final Reviewer verdict.

Evidence Artifacts¶

File	Description
`evidence/EV__TFW_20260830-114238_ASSISTED15.md`	Amended AC-1–AC-10 evidence table and verdict
`evidence/assisted15-verification.log`	Task-local commands, exit statuses and static/no-code verification summary
`evidence/assisted15-fixture-results.json`	Privacy-safe prompt/file, role, update and static-manifest scenario outcomes
`evidence/boundary-and-source-attestation.md`	Product boundary, runtime removal, publication attribution and field pre/post attestation
`evidence/templates/`	Rendered A4/presentation outputs and inspected page captures
`evidence/history-compaction-attestation.md`	Coordinator-created one-time compact-history and tree-equality proof

6. Technical Guidance¶

Amendment A1 is product authority. Do not repair or simplify either deleted runtime; remove it and every dependency instead.
“No-code TFW mechanics” applies to shipped editions/ behavior. Ordinary agent tools and task-local diagnostic scripts may verify files, but may not become product commands or operate user registration/update state.
tfw-identity must be understandable and performable as written by an agent with profile reads, one concise human question when needed, exact target checks and ordinary file creation.
/tfw-update must expose compare → classify → plan → explicit gate → recheck → ordinary file changes → verify → candidate/review. Static JSON records are references, not an executable protocol.
Regenerate release-manifest.json after runtime deletion; it excludes itself and includes maintenance-policy.json. Keep the policy declarative and free of script entry points.
Preserve template TI1 and visual evidence. build_a4.py must not import or call task/workflow/identity/maintenance modules and must not be described as a TFW engine.
Preserve unrelated dirty changes and use exact pathspec staging. Do not alter the external tag or field source.
Executor stops after a clean terminal RF/evidence commit. The Coordinator alone performs the single history compaction at the next safe boundary, then returns the same Reviewer for the terminal full rerun.

7. Definition of Failure¶

❌ Any task-attributed product file outside editions/ changes, including .tfw, a root guide or Light.
❌ Final product exceeds 33 paths, 23 new files or 2,600 changed lines without a new owner amendment.
❌ tfw_identity.py, assisted_maintenance.py or another executable TFW registration/identity/lifecycle/update/maintenance/self-test mechanism remains or is reintroduced under a different name.
❌ build_a4.py performs anything beyond artifact-template building.
❌ Prompt identity guesses a participant/role/owner, overwrites a profile, writes a machine registry or claims OS-level confinement.
❌ Agent-led update silently handles an unresolved path or changed baseline, overwrites downstream state, claims transactionality or skips the explicit gate.
❌ Reverse work mutates public core directly, mirrors the field tree or leaks organization/private context.
❌ Any write reaches the Innoforce source or adjacent versions; any company/brand/history residue enters product or public evidence.
❌ Templates remain branded/placeholders, require network access, accept active resources or lack readable visual evidence.
❌ Product docs/metadata reference removed runtime or contradict prompt/file mechanics, version, authority classes or changelog history.
❌ RF/EV counts stale runtime runs as amended evidence or conceals the concurrent external local tag.
❌ History is compacted before terminal amended RF, compacted more than once, uses a tag as backup, changes the final tree, drops unrelated TFW-55/TFW-60 commits or disturbs dirty files.
❌ Final review is delta-only, uses a new role session, or does not reopen the complete amended contract.
❌ An Assisted task operation pushes, creates/deletes/rewrites a tag or publishes remotely.

8. Phase Risks¶

Risk	Mitigation
Removing runtime leaves stale commands or metadata	Full removed-path/reference scan plus independent semantic review
Prompt identity becomes vague or silently guesses	Closed written scenario table, one-profile rule and explicit question on ambiguity
Static maintenance guidance is too weak to preserve downstream state	Exact authority table, before/after manifests, explicit gate and clean/drifted fixture walkthroughs
“No-code” accidentally removes the useful artifact builder	Keep one explicit exception and semantically inspect `build_a4.py` boundaries
Task-local verification is mistaken for shipped product runtime	Keep it under task evidence only; product references no evidence command
Existing evidence overstates the superseded runtime	Rewrite RF/EV truthfully and count only current amended attachments
Concurrent tag makes publication wording stale	Attribute acts, record current containment, never mutate the external tag
One-time history rewrite disturbs shared work	Use a backup branch, isolated rewrite, exact tree/commit audits and atomic ref update only at a clean task boundary
Unrelated dirty work enters a commit	Exact pathspec staging plus cached-diff and final status audit

TS — TFW_20260830-114238_ASSISTED15: Prompt/File Assisted 1.5 Product and Maintenance Bridge | 2026-08-30