v0.3.3.3 — Second-Order Reflexive Meta-Engineering: Evolved-LAW (Lean · Agile · Waterfall)

IMPLEMENTAUDIT v0.3.3.3

Second-Order Reflexive Meta-Engineering: Evolved-LAW (Lean · Agile · Waterfall)

This release packages:
sustained repository work that
preserves authority, currentness, evidence, proportionality and receiver-reconstructible closure
across planning, implementation, review, recovery and publication.

This release completes the v0.3.3.3 R01–R41 native architecture correction while retaining installed runtime identity 0.3.3.

What changed

• Receiver-complete handoffs (R31).
  A handoff is ready only when it
  carries the implementation state
  and the acceptance/reproduction state
  required by the receiving continuation.
  Ordinary implementation-only handoffs retain the cheap path.
• Work-conserving cell scheduling (R34).
  The runtime distinguishes
  actual dependency and authority conflicts
  from shared issue or milestone labels,
  reconciles governed work into current lifecycle states,
  recomputes the ready frontier after material transitions,
  accounts for active occupancy,
  and honours optional operator resource ceilings.
  Small or strictly serial work does not acquire scheduler ceremony.
• Conditional role separation and capability-fit delegation (R40/R41).
  A separate planner, executor or reviewer
  is selected only when consequence, uncertainty or information value warrants it.
  Delegation is capability-first and least-cost-sufficient,
  with explicit STOP, reconstruction, privacy, context and authority boundaries.
• Observation-bound mutation (R36).
  Mutation requires complete, current and authoritative state for the governed target.
  Governed writers are serialised;
  detected interference fails closed;
  post-effect drift and release uncertainty remain truthful nonterminal residue.
  Portable exclusion of arbitrary non-cooperating same-principal mutation
  remains a target-system security-substrate property,
  not a claim of this runtime.
• Evolved-LAW.
  Source-first Lean, Agile and Waterfall research
  was stripped of unsupported ceremony and crosswalked against native controls.
  Most useful properties were already native;
  the remaining
  proportionality, feedback, protective-slack and consequence-aware assurance refinements
  were absorbed into methodology-neutral IMPLEMENTAUDIT behaviour.
  No methodology runtime or mandatory ceremony was added.
• Optional orientation remains bounded.
  Graphify, ActiveGraph and TokenSave remain optional derived-navigation inputs,
  subordinate to live source and current evidence.
  They do not gain mutation, acceptance or repository authority.

Install or upgrade

Copied skills do not update automatically. From a checkout of this repository:

bash scripts/install-codex-from-release.sh \
  --tag v0.3.3.3 \
  --repo theislampill/IMPLEMENTAUDIT.md \
  --codex-home "$HOME/.codex" \
  --version 0.3.3

Claude Desktop and Claude Code users should follow the host-specific routes in the README.
Copy success is not proof that a host loaded or executed the skill.

Qualification

• Source commit: e7a733be10338a398d0112454088ae3dc2b56f60
• Source tree: 021f1269f081589a66db98f181981e2ad69d1b9a
• Hosted validation: run 31644290216
• Pages build/deployment: run 31644290203
• Installed package: 49/49 payload files byte-identical; runtime 0.3.3

IMPLEMENTAUDIT.skill:

• 257,998 bytes
• 51 archive members
• SHA-256 0e10a21600062c6f7cd440a6efb0ff3b795ce1a01067efb4005c7a176db02413
• 2,002 bytes of headroom under the 260,000-byte calibrated ceiling and outer guard

CHECKSUMS.txt validates the named skill bytes.
It is not a signature, attestation, SBOM or provenance proof.

Evidence limits

This release proves the exact source, package, hosted and installed surfaces recorded for this candidate.
Public release/readback is a separate terminal gate.
The evidence does not establish universal effectiveness across every model, agent, host, repository or external tool.
Graphs and model outputs remain orientation or candidate evidence until independent current-state checks establish the claim they support.
R28/#117 remains optional, owner-paused and nonblocking.
The complete #144 Graphify freshness rebuild is intentionally post-release and will be anchored to the released v0.3.3.3 identity.

Detailed engineering evidence remains in the
v0.3.3.3 release report.

v0.3.3.0 — Evidence fidelity, durable execution, and self-correcting governance

IMPLEMENTAUDIT v0.3.3.0 — Trustworthy evidence, durable runs, and safer releases

v0.3.3.0 is a reliability and hardening release built from sustained use of IMPLEMENTAUDIT across real implementation, review, recovery, integration, and release work.

The release focuses on a simple question:

When IMPLEMENTAUDIT says that something
passed, changed, resumed, merged, published, or closed,
can another person verify that claim from the evidence that survived?

Real campaigns exposed concrete failure modes:
truncated logs could hide failures, shell pipelines could report the wrong exit status,
editing commands could claim success without landing the intended change,
interrupted workers could lose completed work, and remote state could be assumed instead of read back.

v0.3.3.0 turns those lessons into durable rules, checks, and recovery paths.

What changes in day-to-day use

• Verification keeps the complete output and the real producer exit status.
• Interrupted work can resume from durable state instead of replaying an entire programme.
• Long-running work can declare how it finishes, how often it should be checked, and which process or resource it owns.
• Small repairs can use a compact micro-run instead of the full campaign artifact set.
• Independent review records who or what actually performed it, what source state was reviewed, and whether the review was genuinely separate.
• Repeated defects stop being rediscovered indefinitely: they must eventually change the governing rule or acquire a durable owner.

Evidence that can support the claim being made

Verification output is treated as evidence from the moment it is produced.

Checks preserve their complete output and real exit status.
A short transcript excerpt may still be shown for readability, but it is not allowed to replace the full capture or silently change a failing verdict into a passing one.

Claims such as “all,” “none remaining,” and “N of N” must identify the population examined and explain how it was enumerated.
Partial captures, stale snapshots, and instruments that cannot distinguish their inputs cannot prove complete coverage merely because they returned green.

A phase cannot be marked complete when a declared evidence capture is missing or empty.

External state follows the same rule.
The output of a command that edits, pushes, merges, publishes, deploys, comments, or closes something is not proof that the effect landed.
The resulting object—and any automatic workflow or deployment triggered by it—must be read back from the system that owns it.

Changes prove that they actually landed

A successful editing command is no longer enough to establish that the intended change happened.

IMPLEMENTAUDIT can require explicit postconditions,
detect generated files before they are edited,
preserve deliberately duplicated implementations through parity checks,
and account for hashes or identities that a change may invalidate.

Verification also gains a stable window:
the work being checked is not silently changed underneath the running verification.

When one named concept is changed, the run can enumerate every affected
code, documentation,
state, and external-record location
before propagation.
This reduces partial updates in which different parts of the project describe different rules.

Authorisation, interruption, and recovery survive context loss

Owner authorisation is stored as durable run state and bound to its source, rather than depending only on conversation history.

Deferred, blocked, superseded, and non-verdict work remains visible.
Pending terminal actions can be resumed without replaying the entire programme.
Completed child work, partial child work, and genuinely absent work are distinguished instead of being flattened into the same “failed” state.

Provisional output can survive an interruption without being promoted into a verdict.
Owned processes and resources are recorded so recovery does not depend on guessing what a disappeared worker created.

Supported run paths now validate the run root,
and small repairs can use micro-run mode to retain the important evidence and closure rules without the full footprint of a large phased campaign.

Less supervision without weaker checks

Long-running commands can declare
terminal signals, polling limits,
reporting cadence, transport limits,
timeout behavior, kill authority,
and owned process identity.
This reduces repeated “is it done yet?” turns while preserving a clear distinction between running, timed out, interrupted, passed, and failed.

Independent units can be batched when their independence is established,
instead of paying a complete implementation-review-verification cycle for every tiny sibling change.

Verification depth is proportional to the risk being retired.
A proposed repeat verification must identify what changed or what residual risk the extra run will remove.

Before an expensive, scarce, create-once, or externally consequential launch,
the exact wrapper, path, arguments, environment, and terminal-write behavior
can be rehearsed against a safe substitute.

Independent review is now an auditable property

The canonical final lifecycle is:

6 → 6.i → 6.ii → 7

where:

• 6 is plan review and self-critique;
• 6.i is independent cold review;
• 6.ii is preflight;
• 7 is handoff or dispatch.

Legacy 6.2 → 6.i and 6.5 → 6.ii aliases remain readable for compatibility.

Independent review records
the separation, reviewer identity,
requested and resolved runtime/model identity,
reviewed source anchor, and relevant concurrent changes.
A clean verdict can no longer conceal silent model substitution, a stale review anchor, or repeated attempts using effectively the same configuration.

A deterministic refusal caused by the review packet itself is distinguished from a transient transport failure.
Sending the same packet to nominally fresh reviewers is not counted as independent progress when the failure is content-deterministic.

Free-text acceptance checks gain paired positive and negative controls.
Scored prompts cannot contain the answers they are meant to test.
Diagnostic and formal paths need parity when one is used to reason about the other,
and a measuring instrument needs a known-good signal so a broken instrument cannot manufacture a confident false result.

Repeated problems must improve the mechanism

Repeated environment quirks can be recorded and reused instead of rediscovered.

Repeated failures of the same class can mechanically trigger a review of the governing rule, rather than another indistinguishable repair attempt.
A repository defect that appears in a second independent run must acquire a durable tracker or owner, or an explicit recorded refusal.

Countermeasures state what they target, what else they might affect, and how non-target behavior was protected.
Continued repair remains allowed; what changes is that repeated evidence must eventually cause the process itself to be questioned.

IMPLEMENTAUDIT can audit its own process

Process-history and transcript-corpus retrospectives are now first-class audit objects.
They use the same evidence, census, citation, independent-review, residual, and closure rules as implementation work.

An audit of prior audits does not create an infinite hierarchy.
It is another governed audit under the same rules, with an explicit stopping boundary.

Commit history is treated as durable cross-context evidence.
Meaningful commits can record
what changed, why it changed,
the evidence anchor, the relevant boundaries,
and the finding or decision that motivated the change.

The audit playbook also gives operational proof standards for
error handling, type-strength boundaries,
duplication, dependency cycles, and dead-code claims.

Evaluator, validation, and Linux hardening

The v0.3.3.0 campaign applied the new code-hygiene rules to IMPLEMENTAUDIT’s own evaluator.

Broad exception handling and failure-reporting paths were censused.
Sites where failures could be hidden behind normal-looking results were repaired,
while legitimate fail-closed and best-effort behavior was retained and documented.

Hosted Linux validation was repaired and requalified.
The release campaign also removed a platform-specific fixture-cleanup failure,
provisioned the tools and Git history required by the tests,
and removed a timing race from child-process recovery testing.

Validation registry checks no longer rely on a false documentation-test exemption or brittle hard-coded suite counts.
Adding a legitimate test should not create a misleading release failure,
and an unregistered validation path should not silently disappear from the complete verifier.

Safer publication, integration, and release identity

A version label is no longer treated as sufficient package identity.
The source anchor, package bytes, and digest identify the release together.
Replacing bytes under an existing version requires an explicit superseded-to-superseding digest record,
and stale build artifacts cannot silently stand in for the current source.

External publication now has a governed path:
establish the intended scope, check for overlap,
make citations and source references resolvable,
obtain the required independent review and authorisation,
publish, and then read the published object back.

Pushes and merges declare the workflows, deployments, issue transitions, and other automatic effects they are expected to trigger.
Those effects are read back after the mutation.

Multi-issue campaigns declare their branch and integration topology before implementation.
Stacked integration can verify
retargeting, patch identity,
synthetic merge results, exact heads,
ancestry, and resulti...

v0.3.2.0 — Evidence integrity, failure-origin discipline, and context continuity

IMPLEMENTAUDIT v0.3.2.0 — Evidence integrity, failure-origin discipline, and context continuity

v0.3.1.0 changed how IMPLEMENTAUDIT’s guidance is packaged and loaded. This corrected v0.3.2.0 re-release builds on that progressive-disclosure runtime and changes how IMPLEMENTAUDIT identifies evidence, attributes failures, preserves continuity, governs closure, and validates those behaviors under real model execution.

Highlights

Stronger evidence integrity and custody

Evidence can now be bound to the full repository revision that produced it and identified as structural, behavioral, or provenance evidence. Closure claims identify the surface actually verified—such as source, generated output, package, installed copy, running system, deployment, API, or publication—so lower-layer success is not promoted into proof of a higher layer.

Formal evaluation paths also preserve stronger process, checkout, session, raw-output, and artifact custody. Missing or substituted evidence fails closed. Interrupted long-running work records explicit states such as terminal-state-unverified or infrastructure-failed rather than guessing success, guessing failure, or silently replacing the run.

Failure-origin and closure discipline

IMPLEMENTAUDIT now more clearly distinguishes product defects from rule or validator defects, host and infrastructure failures, evidence or custody failures, authorization problems, and genuinely unresolved states.

Multiple defects associated with one occurrence remain separately visible. Consequential residuals require an explicit disposition, and unresolved causes are not promoted into complete root-cause claims. A safe containment route can be recorded without pretending that every cause has been resolved.

Repeated abnormalities can also trigger review of the governing rule itself. A passing validator, scorer, or evidence standard is no longer assumed adequate merely because it returned green.

Continuity, authorization, and recurrence prevention

After a real context boundary—such as compaction, a new session, or a handoff resume—the receiving session rereads live state before mutation. Git and durable run evidence win over reconstructed summaries. Already-satisfied one-time work is not silently replayed, while standing constraints and authorizations remain in force.

Consequential actions are bound to the parameters the owner actually authorized. Missing or conflicting parameters raise authority drift instead of adopting tool defaults.

Recurring lessons can be lifted into durable guidance, but closure distinguishes writing an encoding, making it mechanically active, installing the current version, and later proving recurrence prevention. Completion markers are also emitted once at their real transition and are not replayed in later summaries.

Real model-in-the-loop qualification

The frozen candidate passed the defined fourteen-cell behavioral matrix under both Luna and Opus:

• Luna Matrix: 14/14
• Opus Matrix: 14/14

The matrix exercises governed execution, evidence and transcript discipline, failure-origin reasoning, residual handling, authorization, claim verification, second-order review, lesson activation, enumeration, and related boundary behavior.

The supplementary B3 continuity campaign also passed under both models:

• Luna B3: 6/6
• Opus B3: 6/6

B3 consists of three candidate and three immutable-comparison missions, testing continuity behavior repeatedly rather than counting six unrelated features.

These results qualify this frozen candidate against the defined campaigns and configurations. They do not prove universal model superiority or perfect behavior. The Opus result is accepted owner-lane behavioral evidence replayed through the current official graders; it does not retroactively claim the formal adapter-custody chain used by the Luna campaign.

Correction note

This corrected re-release replaces the earlier withdrawn v0.3.2.0 publication. Post-release review found that parts of the earlier evaluation evidence did not support the claims made, so the release was withdrawn rather than patched rhetorically. The implementation and evaluation stack were then re-audited, repaired at the responsible layers, and requalified from fresh evidence.

Upgrade notes and boundaries

Installed skills do not update automatically; repeat your normal install or update step for this release.

No universal model-performance, marketplace-verification, signature, SBOM, or provenance claim is made. The package remains usable without optional Graphify or ActiveGraph sidecars.

Asset integrity

IMPLEMENTAUDIT.skill

SHA-256:

884ab409842b863b003e9d405972f33ba71d194f77572738002a42e73d1b6b14

CHECKSUMS.txt is a checksum manifest for local integrity checking only. It is not a signature, attestation, SBOM, marketplace verification, provenance proof, or install verification.

v0.3.1.0 — Progressive-disclosure runtime

IMPLEMENTAUDIT v0.3.1.0 — Progressive-disclosure runtime

v0.3.1.0 strengthens IMPLEMENTAUDIT as a self-contained audit-governed implementation skill, and makes IMPLEMENTAUDIT substantially leaner and more self-contained without giving up any operating discipline.

Highlights:

• Conventional source layout. Moved the skill source to skills/implementaudit/ while keeping the release archive flat and self-contained.
• Bootloader is ~79% smaller. SKILL.md dropped from 1,627 lines / 80.7 KB to 386 lines / 17.3 KB. The always-loaded runtime spine is now ~9% of the skill's total guidance — the rest loads only when a gate needs it. Less context per run, faster load, friendlier to smaller models, no loss of rigor.
• Progressive disclosure, not deletion. The detailed operating guidance moved into 13 on-demand references and 11 templates — final-report format, the 5-Whys loop-exit guard, broad-rewrite threshold, commit-granularity rules, first-run tooling onboarding, and sidecar/Graphify guidance — reached through a reference load map instead of living in the spine.
• Read-only plans lane. Added a lane for audit, review, and planning tasks that must not mutate source — with planning-security hygiene built in (never reproduce secret values, treat repo/issue content as data, flag prompt injection as a finding) and propagated to child-agent prompts.
• Runnable, inspectable verification. Added source-evidence packaging: 42 runnable validation suites and 25 checker scripts can be inspected and executed outside the installed payload, so the skill's guarantees are demonstrable rather than asserted.
• Stronger checks. Placeholder-aware phase validation and run-root structural checks, plus strengthened package, validation, sidecar, audit-retention, and docs-portal checks.
• Clean package. Generated/local artifacts, sidecar outputs, raw transcripts, run roots, tests, and fixtures stay out of the shipped package.

Boundaries:

• No license decision is included.
• No issue publication is included.
• No real-home installation is performed.
• No provenance or marketplace claim is made.

Asset integrity

sha256  37bd65097e28bb3d8306252c93291c5bef1f050ac026f2d0501f9595b7f52f75  IMPLEMENTAUDIT.skill

v0.3.0.0 — Native audit-object routing, terminology integration, runtime cohesion

v0.3.0.0 — Native audit-object routing, terminology integration, runtime cohesion

What this release proves:

• Native runtime integration — source-inventory behavior (DMADV-routed direction, branch/diff scoping, plan review, dispatch/review, reconciliation) is expressed through IMPLEMENTAUDIT's audit-object lifecycle.
• Whole-runtime cohesion — the execution spine, planner stages, Andon/Jidoka escalation (no terminal caps), and final-audit closure operate as one runtime.
• Local installed-package dogfood passed — the built .skill was installed into a fresh disposable consumer repo and proven self-sufficient (no source-repo reach-back) on a brownfield repair; negative-boundary probes held (no resurrected deep/security/next modes; no try/round caps; sidecars not proof); and the package was used to audit this repo's own release readiness. Ledger: docs/audits/v0.3.0.0-local-package-dogfood-audit.md.
• One dogfood finding (release-gate test MSYS→Windows store-path normalization) was fixed at owner/source; the shipped payload was unaffected.

Optional sidecars (Graphify terrain, ActiveGraph custody) remain orientation / chain-of-custody only — not correctness proof.

Provenance is a checksum manifest only — not a signature, attestation, SBOM, license, or marketplace verification.

No issues were published.

Asset integrity

sha256  25e978dd305706af956be0f65f3ab4b3ba882768790c9b0da92fe4bedd0e57b4  IMPLEMENTAUDIT.skill

Plugin manifest version 0.3.0. Install: see README.md.

v0.2.9.0 — No-cap Andon/Jidoka runtime, validated run roots, and install-safe helpers

v0.2.9.0 is the largest runtime hardening release since the v0.2.4.x line.

A proper Andon/Jidoka escalation contract:

ANDON_PROBE -> ANDON_ESCALATE -> ANDON_HANDOFF

There are no arbitrary try caps or final-audit round caps.

ANDON_PROBE runs on the first abnormality and records proportional 5 Whys, Hansei, containment, countermeasure, and required rerun evidence.

ANDON_ESCALATE requires actual progress evidence — New evidence: and/or Changed approach: — and must cite prior same-class Andon log rows before treating a recurrence as the same class.

ANDON_HANDOFF is reserved for real blockers: unsafe scope, missing authorization, owner decision, external dependency, irreproducibility, missing access/tooling, or no bounded countermeasure remaining.

What changed

• No-cap Andon/Jidoka runtime

  • Added classed Andon logging to STATE.md.
  • Promoted official abnormality classes into the transcript contract.
  • Added recovered and blocked Andon transcript skeletons.
  • Added checker coverage so capped-retry language and malformed escalation transcripts cannot quietly return.

• Validated run-root substrate

  • Added validate-run-root.sh for dispatched run-root conformance and resume safety.
  • Completed shipped run-root templates for state, protocol, tools, context, and sidecars.
  • Tightened status vocabulary, pause/continuity marker handling, AUDIT_GAPS ordering, and final-audit transcript rules.
  • Isolated .IMPLEMENTAUDIT/ run-root artifacts from target-repo evidence scans so local audit debris does not contaminate cleanliness checks.

• Installed-consumer helper repair

  • Added IMPLEMENTAUDIT_SKILL_DIR helper-path resolution so packaged helper calls work from installed skills as well as from the source repo.
  • Hardened shipped helper scripts with smoke tests in minimal target repos.
  • Repaired helper-path and variable-collision issues found during dogfooding.

• Package and validation hardening

  • Package manifests now use stricter set-equality checks.
  • Shipped-path integrity checks prevent repo-only paths from leaking into installed payload instructions.
  • Validation registry parity now keeps test coverage wired into both local validation and CI.
  • README ToC, install/release claims, portal claims, and validation lists were reconciled against the live release state.

• Agent-eval and transcript grading

  • Added an adversarial fixtures/agent-eval/ pack for common misreads:
    terminal retry caps, autonomous-builder overreach, audit-only misunderstanding, release-bot overreach, and Lean-glossary theater.
  • Added a deterministic transcript grader for required markers, forbidden phrases, marker order, and no-terminal-cap checks.
  • These fixtures are evaluation inputs and expected transcript properties, not proof of live model behavior by themselves.

• Optional sidecar conventions

  • Graphify and ActiveGraph remain optional for consumer repos.
  • Markdown fallback remains first-class.
  • For maintaining IMPLEMENTAUDIT itself, the repo now has clearer sidecar conventions: per-run custody stores, absent-safe custody-append.sh, caused-by chaining for Andon custody events, and V0260-labeled historical backfill rules.
  • Sidecar evidence remains custody/orientation evidence, not correctness proof.

Upgrade notes

Installed skills do not auto-update. Repeat your install/update step for this release.

CHECKSUMS.txt is a SHA-256 checksum manifest for local integrity checking only. It is not a signature, attestation, SBOM, marketplace verification, provenance claim, or install verification.

No consumer repo is required to install Graphify or ActiveGraph. The skill remains usable with Markdown evidence and ordinary live-file Gemba.

v0.2.8.0 — Adaptation lane: casual-build intake, bounded continuity preload

IMPLEMENTAUDIT v0.2.8.0

Adds governed casual-build intake, bounded continuity preload with IMPLEMENTAUDIT_CONTINUITY_SAVED marker, optional Polish & Harden phase shape, a docs portal generator, and GitHub Pages CI.

What changed

G1 — Chooser table: skills/references/goal-format.md now has a 4-row invocation chooser (normal prompt / /goal / /implementaudit / governed casual-build intake) with easy invocation examples.

G2 — Neutral identity hygiene: docs/portal/onboarding.md §Compared uses neutral comparator language only. Formal runtime forbidden-term check passed (0 occurrences in all native surfaces).

G3 — Governed casual-build intake: Added as the 4th invocation shape in skills/SKILL.md §2/§2a/Stage 0. Natural-language intent is synthesized into a tdqyq-audit-object before any mutation. Covered by fixtures/casual-build/.

G4 — Bounded continuity preload: skills/SKILL.md Stage 0 now defines a 5-source priority order (AGENTS.md → run-root note → personal/project notes → Graphify terrain → ActiveGraph custody) and the IMPLEMENTAUDIT_CONTINUITY_SAVED marker with 6 required fields.

G5 — Per-phase continuity writeback (STRENGTHENED): v0.2.6.0 PROTOCOL loop was the base. v0.2.8.0 adds the IMPLEMENTAUDIT_CONTINUITY_SAVED marker, bounded 5-option writeback table in PROTOCOL.md Steps 14/15, ActiveGraph custody path, Graphify terrain-update request, and 9 new continuity checks (34/34 total).

G6 — Polish & Harden phase shape: Rule P4-8 in skills/references/phase-design.md. Optional terminal phase; no new features; fixtures in fixtures/phase-design/polish-harden.md.

G7 — Simpler ship-it story: scripts/build-docs-portal.py (stdlib-only portal generator), scripts/check-docs-portal.py (12-check validator), tests/docs-portal.test.sh (26/26), docs/portal/onboarding.md (10 onboarding sections), .github/workflows/pages.yml (GitHub Pages CI).

Audit-fix round (5 items before AUDIT_COMPLETE):

• Graphify authorization boundary reconciled (terrain reads authorized as self-maintenance; no new indexing)
• G5 reclassified from ALREADY COVERED to STRENGTHENED
• Python print() replaced with sys.stdout.write/sys.stderr.write in new CLI scripts; no broad *.py cleanliness exemption
• Workflow structural checker (scripts/check-workflow-structure.py) added; full schema validation pending CI
• Formal identity hygiene check: 0 comparator terms in all native surfaces

Install

Codex: bash install-from-release.sh or extract IMPLEMENTAUDIT.skill to ~/skills/implementaudit/.

Claude Desktop: import IMPLEMENTAUDIT.skill via the import UI, then restart Claude Desktop.

Checksum

sha256  718075c053f5a5341168d9697d85ce4d95349dffb8b77885f3a142cabe3ec434  IMPLEMENTAUDIT.skill

Claim boundaries

• Checksum manifest only — CHECKSUMS.txt is a SHA-256 checksum file. It is not a code signature, attestation, SBOM, provenance claim, or marketplace verification.
• No marketplace publication — No skill marketplace submission, update, or publication was performed for this release.
• No install proof — No live Claude Desktop or Codex install was performed for this release. File-copy smoke test only.
• Graphify / ActiveGraph optional — Both sidecars are optional and absent-safe for user runs. For IMPLEMENTAUDIT self-maintenance runs, Graphify and ActiveGraph are canonical (since v0.2.4.5) but do not constitute correctness proof by themselves.
• Docs portal — Published via GitHub Pages. URL: https://theislampill.github.io/IMPLEMENTAUDIT.md/ — availability depends on GitHub Pages propagation.

See docs/audits/v0.2.8.0-adaptation.md for the full audit ledger, ActiveGraph custody chain (19 events), and remaining risk register.

v0.2.7.0 — Lean/TPS operating discipline

What's new

IMPLEMENTAUDIT v0.2.7.0 implements Lean/TPS operating discipline as native runtime behavior across 20 concepts. Lean terms are not decorative labels — each maps to an auditable runtime behavior, owner/source file, and verification check.

New runtime behaviors

• DMAIC brownfield routing (Define→Measure→Analyze→Improve→Control) for defects, regressions, release repairs
• DMADV greenfield routing (Define→Measure→Analyze→Design→Verify) for new governed artifacts
• 5S_CHECK per-phase gate — all five pillars (Seiri/Seiton/Seiso/Seiketsu/Shitsuke) recorded as clean/deferred/blocked at every phase boundary in PROTOCOL.md Step 9
• Jidoka stop-the-line chain — 8-step: Andon → FAILURE_PROBE → Hansei → 5 Whys → countermeasure → Kaizen standardization → re-run evidence → close/block/defer/handoff
• Nemawashi owner-decision gate — surface consequential assumptions before dispatch for release/package/AGENTS.md/sidecar-status changes
• Muda/Mura/Muri register in THINKING.md — waste logged with disposition, not silently absorbed
• Quality route field in phase-goal.txt — DMAIC / DMADV / PDCA / 5S / mixed / not applicable
• check-lean-discipline.sh poka-yoke checker — 12 structural requirements, verified by tests/lean-discipline.test.sh 11/11

Sidecar gate

Graphify terrain extraction ran live (177 nodes / 202 edges across 6 corpora). Native semantic extraction via the claude-cli backend succeeded using a Claude Max subscription — no separate Anthropic API key required. Graphify is terrain/orientation evidence only; all architecture candidates confirmed against live files. ActiveGraph live custody: 8 events written and read back. Inter-leverage proven on an isolated target-repo fixture. No sidecar outputs committed or included in the .skill package.

Evidence boundaries

No Lean/TPS certification, sigma level, DPMO, or statistical process control values are claimed. DMAIC/DMADV are routing and evidence-shaping patterns for audit-governed repo work, not Six Sigma certification claims.

Install

See README.md for full install instructions.

Checksums

sha256  19b4f41aa6feb423cb1bdaf9be6de719b211805b2ecdf2dc7b3ba4fe45c32c86  IMPLEMENTAUDIT.skill

Full changelog: CHANGELOG.md

v0.2.6.0 — Executable Phase Specs, Strict Validators, and Sidecar-Custody Release Evidence

What's in v0.2.6.0

Executable phase specs

skills/templates/phase-goal.txt is now a full executable spec with all required structural fields — Phase N/TOTAL, Task, Type, Run root, Baseline ref, Owner/source, Audit object, Auditing operation, Terminal object state, Mandatory commands, Acceptance criteria, Evidence required, Depends on, Why, Work, IMPLEMENTAUDIT_PHASE_VERIFY, AGENTS_UPDATE_DECISION, CONTINUITY_DECISION, and IMPLEMENTAUDIT_PHASE_DONE. Placeholder-only acceptance criteria are now a validation failure.

Strict validators

• validate-phase.sh: 19 failure-mode checks, up from placeholder-only. Wired into Stage 5 dispatch-prep and the per-phase execution loop.
• check-forbidden-terms.sh: generic caller-supplied forbidden-term gate; the runtime-provided term is never embedded in source. Used at every release gate.
• phase-validation.test.sh: 20/20 (1 positive + 19 failure modes).

Concrete protocol loop

skills/templates/PROTOCOL.md now contains:

• 16-step per-phase execution loop (read STATE.md → validate spec → Smoke A → execute → mandatory commands → acceptance criteria → cleanliness check → Smoke B → IMPLEMENTAUDIT_PHASE_VERIFY → AGENTS_UPDATE_DECISION → CONTINUITY_DECISION → IMPLEMENTAUDIT_PHASE_DONE)
• Exact 3-strike failure recovery ladder: FAILURE_PROBE → FAILURE_ESCALATE → FAILURE_HANDOFF
• 3-round final audit protocol with coverage math and trust-prior > 30% warning
• Mid-run interruption handling: IMPLEMENTAUDIT_PAUSE, 4-option menu (Resume / Revise spec / Skip phase / Stop)
• Ordering rules: AUDIT_COMPLETE must precede IMPLEMENTAUDIT_RUN_COMPLETE

Sidecar-custody release evidence

• Graphify 0.8.35 used for project terrain orientation. Output is orientation evidence, not proof. Graphify outputs are excluded from tracked source and the .skill package.
• ActiveGraph 1.0.5.post2 used for live release-gate custody. 11 live events recorded with custody_mode: live_release_gate.
• Historical backfill: 26 events covering v0.2.4.5, v0.2.5.0, and v0.2.6.0 phases 1–14 are marked custody_mode: historical_backfill with source, backfilled_at, original_event_time, and evidence_boundary fields. Backfill is transcript-derived evidence, not live gate passage. ActiveGraph custody is chain-of-custody evidence only, not correctness proof.

Package boundary

The .skill asset contains only runtime skill payload and plugin metadata. Repo docs, tests, fixtures, release scripts, audit ledgers, run roots, and sidecar stores are excluded. A 23-entry package boundary manifest with 14 proven-excluded categories is recorded in docs/audits/v0.2.6.0-final-runtime-and-package-boundary-audit.md.

Asset compression: A ZIP compression bug has been fixed. All 23 entries are now ZIP_DEFLATED. Asset size: 59,551 bytes (was 154,750 bytes). Compression regression is caught automatically by tests/release-asset.test.sh on every run.

Capability ledger entries

• SIDECAR-OPTIONAL-V0260: Graphify and ActiveGraph sidecar dogfood PASS — real install, real project terrain/custody evidence, real boundary enforcement. Absence remains non-blocking.
• HISTORICAL-BACKFILL-V0260: ActiveGraph historical backfill gate PASS — 26 events, all custody_mode: historical_backfill.

Test results

Suite	Result
phase-validation.test.sh	20/20
continuity.test.sh	25/25
sidecars.test.sh	24/24
release-asset.test.sh (incl. compression check)	ok
release-asset-install.test.sh	ok
capability-ledger.test.sh	ok
verify-package.sh	ok

Post-release documentation repair (32b9642)

Commit 32b9642 closes post-release documentation Andons. The .skill asset and checksum are unchanged; no new version is declared.

• execution-spine.mmd — adds PhasedRun subgraph documenting the v0.2.6.0 phased execution runtime: run-root claim → Stage 6 plan review → Stage 6.5 preflight smoke → Stage 7 handoff → validate-phase.sh → 16-step phase loop → 3-strike failure recovery → final audit + audit-fix rounds → AUDIT_COMPLETE.
• invocation-modes.mmd — updates the Synthesis artifacts node to show the namespaced run-root structure (ROADMAP · STATE · THINKING · PROTOCOL · sidecars · phases/phase-N.md).
• README.md — updates run-root prose to reference .IMPLEMENTAUDIT/runs/<task-slug>-<id>/phases/ with legacy-compat note; adds explicit .skill archive boundary statement; updates project milestone and plugin manifest version to v0.2.6.0.
• v0.2.6.0-final-runtime-and-package-boundary-audit.md — corrects proof matrix summary counts: native-owner-covered 15 → 18; covered-with-stronger-implementaudit-boundary 9 → 7.

Checksums

sha256  f37f6356b3a342510512e57c3f2785b6be4ea044a5b312d59d40d8ce4eade947  IMPLEMENTAUDIT.skill

This is a checksum manifest only. It is not a signature, attestation, SBOM, license, marketplace verification, or universal host support claim.

v0.2.5.0 — Staged-goal behaviors (repaired for Claude install)

v0.2.5.0 - Staged-goal behaviors (repaired for Claude install)

v0.2.5.0 adapts useful staged-goal behaviors into runtime contracts.

Highlights

• Strengthened Stage 0 context/tool/memory/sidecar detection.
• Strengthened Stage 1 routing with full greenfield category intake and brownfield 0-2 true-gap questions after recon.
• Added namespaced run-root claiming with .IMPLEMENTAUDIT/runs/<task-slug>-<id>/ and skills/scripts/claim-run.sh.
• Strengthened Stage 6 self-critique/review menu and Stage 6.5 red preflight discipline.
• Added bounded continuity via CONTINUITY_DECISION.
• Strengthened Graphify and ActiveGraph boundaries: optional, noncanonical, absent-safe, and excluded from package debris.
• Strengthened Capability Ledger wording so entries are derived from recorded evidence, not broad competence claims.

Release assets

• IMPLEMENTAUDIT.skill SHA256: 53b56a5dba67263ef9648289980982b75d0b037acc6ece41811346e9fe425566
• CHECKSUMS.txt — bounded artifact-integrity evidence for the uploaded .skill asset.
• The checksum manifest is not a signature, attestation, SBOM, license, marketplace verification, universal host support, passive update mechanism, or import proof.

Archive root shape

The .skill archive is structured for Claude Desktop import:

IMPLEMENTAUDIT.skill (ZIP):
  SKILL.md              ← at archive root
  references/
  scripts/
  templates/
  .claude-plugin/
    plugin.json
    marketplace.json

Validation (repair HEAD: 0a42476)

• GitHub Actions validate passed for repair HEAD 0a42476.
• Local package verification passed with bash scripts/verify-package.sh.
• Release asset build validation passed including root-shape regression guard.
• Codex install smoke passed (stale checksum, overbroad archive, sidecar-output rejection).
• Claude archive smoke passed (same canaries, wrong-root-shape rejection added).
• Live public v0.2.5.0 asset redownloaded and verified against CHECKSUMS.txt.

Boundaries

• No root IMPLEMENTAUDIT.md behavior file is included or recreated.
• No marketplace verification, universal host support, license claim, passive update automation, signature, attestation, SBOM, or broader provenance claim is made.

---

Post-release repair (2026-06-07): LIVE_V0_2_5_0_CLAUDE_INSTALL_BROKEN

Primary root cause: The original .skill archive had the wrong root shape.
Skill content was nested under skills/SKILL.md but Claude Desktop import requires
SKILL.md at archive root. The archive was accepted by the Codex installer (which
reads from skills/ explicitly) but rejected by Claude Desktop with "malformed at root."
No live Claude Desktop import was run in the original release gate.

Secondary root cause: No Claude Desktop install/import path was documented,
scripted, or smoke-tested.

What was fixed (repair HEAD: 0a42476)

Archive root shape (primary fix):

• scripts/build-release-asset.sh: strips skills/ prefix; SKILL.md now at archive root.
• .claude-plugin/plugin.json: skills field changed from "./skills/" to "./".
• scripts/install-codex-from-release.sh: updated for new root shape.
• scripts/install-claude-from-release.sh: updated for new root shape; relabeled as
  "file-copy workaround only — NOT Claude import proof."
• tests/release-asset.test.sh: regression guard added (skills/SKILL.md in archive = hard failure).
• scripts/verify-package.sh: plugin skills path check updated.

Claude install coverage (secondary fix):

• Added scripts/install-claude-from-release.sh (file-copy workaround, labeled as NOT import proof).
• Added tests/release-asset-install-claude.test.sh for Claude archive shape smoke.
• Added Claude Desktop install path and boundaries to README.md §Install notes.
• Added anti-repeat rule LIVE_V0_2_5_0_CLAUDE_INSTALL_BROKEN to AGENTS.md.

Tag repair

• v0.2.5.0 tag moved to repair HEAD 0a42476 (new tag object d641275...).
• GitHub source archives now contain the fixed builder, tests, and archive validator.
• Original tag 8df3c07c and all intermediate repair commits documented in audit ledger.

Archive SHA

• Repaired archive SHA256: 53b56a5dba67263ef9648289980982b75d0b037acc6ece41811346e9fe425566
• Changed from original a4d953d... due to structural root-shape fix (not platform artifact).

Boundaries after repair

• Live Claude Desktop import proof: pending manual verification after Claude Desktop reload.
  The archive root shape is now correct for Claude import. Verify by downloading
  IMPLEMENTAUDIT.skill from this release and importing via Claude Desktop's skill import UI.
• scripts/install-claude-from-release.sh is a file-copy workaround only.
  It is NOT proof that Claude Desktop can import the .skill file.
• Dogfood of /implementaudit from v0.2.5.0 release asset requires Claude Desktop restart
  after import.

Audit ledger: docs/audits/v0.2.5.0-claude-install-repair.md