skein v0.2.4

[0.2.4] - 2026-06-21

Added

• Mermaid diagrams render live in /plan-view HTML. render_markdown now emits <pre class="mermaid"> for ```mermaid fences (HTML-escaped, <pre> chosen so the raw diagram source stays readable as fallback), and both template.html / plan-template.html load the Mermaid v11 ESM runtime from jsDelivr (mermaid@11, startOnLoad: true, securityLevel: 'strict'). CDN-only by design (no vendoring); diagrams need a network fetch to render. Applied identically to the skein-codex mirror.
• Conditional ## Architecture & Call Flow section in the dev-plan template. Included only when a plan has 2+ independently-executing components: a component graph (Mermaid), a trigger-order sequenceDiagram, and a context-lifecycle table. dev-plan SKILL.md documents the inclusion heuristic, a pre-phase user-confirmation gate, the three sub-elements, placement (after Integration Seams, before Testing Notes), and that the section is above the review marker (immutable contract).
• Negative-space topology audit in the /review-plan architecture lens. The lens now reconstructs the implied call-flow from Files-to-Modify + Technical Specifications and flags components/triggers/context-transitions a plan needs but never names, plus a topology-omission backstop (Missing Task / Important when 2+ components lack an ## Architecture & Call Flow section).
• Interactive triage-and-clarify elicitation loop in /review-plan (default-on Step 6.4): triage via free-form selection over unbounded findings, per-finding clarify options, then route — act-on decisions to /dev-plan update (above the marker), waived findings to a ### Review Waivers subheading (below the marker). New --batch flag skips the loop for unattended/CI runs and is orthogonal to --auto-fix=trivial. A write-then-hash ordering invariant pins: loop edits flush → auto-fix → marker hashed exactly once.
• Model / thinking-level routing table in the skills-architecture design doc (per skill → lens → Claude model+effort → Codex reasoning level) with a maintenance note.

Changed

• Bumped both plugin manifests to 0.2.4 in lockstep (the version field is the update cache key; both mirrors must move together).
• ## Architecture & Call Flow added to AF_FORBIDDEN_HEADINGS (canonical scripts/lib/auto-fix-common.sh, re-bundled to all skill subtrees) and to both review-plan prose forbid lists, so --auto-fix=trivial cannot rewrite the new immutable section.

skein v0.2.3

Fixed

• /review-plan now writes its review marker via a bundled deterministic entrypoint (scripts/write-review-marker.py) instead of an LLM hand-following a prose recipe at Step 7. Root cause (confirmed from session transcripts, both /review-plan and /conduct on 0.2.2): the 0.2.2 fix hardened conduct/marker.py (code) and both SKILL.md mirrors (prose), but /review-plan still computed the marker hash by hand — and an agent, despite the byte-faithful prose in context, wrote above = b'\n'.join(lines[:idx]), which drops the newline immediately above the marker (join separates, it does not terminate). That rstripped hash diverged from conduct's byte-faithful slice (plan_text[:span_start]), so /conduct false-flagged a freshly reviewed plan as drifted. Prose cannot prevent the splitlines/join newline-eating trap; only shared deterministic code can.
• The marker hash is now computed by the same marker.py authority /conduct validates with — promoted to a canonical scripts/marker.py and fanned into both review-plan mirrors by scripts/bundle-appliers.sh (per-skill extras via bundle_extra_for). All copies (conduct ×2, review-plan ×2) are anchored byte-identical to the canonical file by tests/parity/test-marker-parity.sh, so the two skills can no longer disagree on a plan's hash.
• /review-plan Step 7 adopts the auto-fix applier's abort-if-absent / never-hand-compute contract, and now aborts (rather than appending the marker at EOF) when a plan has no marker line and no template placeholder divider — burying the marker below the workspace would fold the workspace into the hashed contract.

Changed

• Bundle map (scripts/lib/bundle-map.sh) grows a per-skill extras hook (bundle_extra_for) so marker.py + write-review-marker.py ship into review-plan only (not deep-review), preserving the "bundled == operative" invariant. check-sync.sh and tests/parity/test-applier-bundle-parity.sh enumerate the extras so their byte-identity is guarded.
• Added tests/parity/test-marker-parity.sh (anchored byte-identity + post-write round-trip + the named 9fa0989-vs-df8d891 divergence regression) and tests/auto-fix/test-review-plan-marker-write.sh (recipe-removed negative assertion across both mirrors, abort contract, placeholder happy path).

skein v0.2.2

Fixed

• Quoted SKILL.md frontmatter values that contain multiple bracketed argument tokens or colon-bearing prose so Codex can load all bundled skein:* skills without YAML parse warnings.
• /conduct review-marker hashing (conduct/marker.py) is now byte-faithful, matching the documented recipe (git hash-object of the bytes above the marker line). strip_marker_for_hashing previously popped the blank line that normally precedes the marker and round-tripped through splitlines()/join(), silently normalizing CRLF→LF; compute_plan_hash also read via read_text, translating line endings before hashing. Together these made the computed hash disagree with git hash-object of the above-marker bytes whenever a plan had a blank line before its marker (the common markdown style) or used CRLF endings — so preflight false-flagged a valid plan as stale (hard-stopping a first run, or, on --resume, rewriting the marker into a form no byte-faithful external checker accepts). Hashing now slices the plan at the marker line's exact byte offset with line endings preserved. Migration: markers written by /review-plan with a blank line before them now validate correctly; any marker previously written by the old marker.py blank-popping path on such a plan reports stale once and refreshes on the next /review-plan or --resume.
• write_marker now hashes the exact bytes it writes above the marker. It previously hashed the contract before appending the trailing newline that separates the contract from the marker line, so marking a plan that had no trailing newline (e.g. a freshly authored plan) recorded a hash of bytes that were never written above the marker — leaving the plan stale the instant /review-plan marked it, and /conduct would then reject a plan the user had just reviewed. (Pre-existing; surfaced by adversarial review.)
• write_marker reads and writes plan bytes directly (read_bytes/write_bytes) instead of text-mode read_text/write_text. Text mode applies universal-newline translation whose direction is platform-dependent — on Windows write_text re-expands \n to \r\n on disk — so the bytes write_marker hashed could differ from the bytes compute_plan_hash (byte-faithful since this release) reads back, marking a CRLF plan stale on Windows the instant it was written. Reading and writing raw bytes keeps both sides on identical bytes on every platform and preserves a plan's existing line endings.

Changed

• Hardened the review-marker locators: last_marker_index (line index) and _marker_line_span (byte offset) now share one _scan_marker_lines fence-tracking core so they can never drift to different markers on the same plan. Documented the byte-faithful invariant in conduct and review-plan SKILL.md (both mirrors) to prevent re-introducing line-ending/blank-line normalization. Added tests/parity/test-conduct-marker-parity.sh (wired into just parity-tests) asserting conduct/marker.py stays byte-identical across the skein and skein-codex mirrors.

Full changelog: CHANGELOG.md

skein v0.2.1

Documentation

• Clarified in both deep-review SKILL.md copies that scripts/render-reconciled-report.sh is a repo-only reference renderer, deliberately not bundled into the installed skill (per scripts/lib/bundle-map.sh) — the running review renders by hand from the report template, so the renderer's absence under the installed scripts/ is expected, not a broken bundle. Prevents review runs from false-flagging it as a missing artifact.

Full changelog: CHANGELOG.md

skein v0.2.0

Added

• Fifth assumptions lens (Opus) for /review-plan: audits claims the plan states as settled fact but cannot verify from the codebase — backend/external behaviour, business semantics, data shape, unread contracts, environmental facts. /review-plan now runs four high-reasoning Opus lenses plus one Haiku factual lens.

Fixed

• Cross-lens reconciliation (scripts/reconcile-findings.sh) no longer collapses unanchored findings. Findings with no location anchor (empty file + -1 line) each receive a unique merge signature instead of sharing one ("", -1, category) signature and silently dropping all but one. Anchored merge/related behaviour is byte-for-byte unchanged; both the jq and awk-fallback parsers partition identically (including an explicit line:-1). Also benefits /deep-review, which shares the engine.

skein v0.1.0 — first public release

First public release of skein — a namespaced skill plugin for Claude Code and the OpenAI Codex CLI. All skills load under the skein: namespace to avoid collisions.

Install

Claude Code:

/plugin marketplace add vr000m/skein
/plugin install skein@skein

Codex CLI:

codex plugin marketplace add vr000m/skein --ref v0.1.0
codex plugin add skein@skein

Added

• Eleven namespaced skills under skein:* shipped as a single plugin for Claude Code and OpenAI Codex CLI: dev-plan, review-plan, conduct, fan-out, deep-review, content-draft, content-review, rfc-finder, spec-compliance, update-docs, plan-view.
• Dual-harness plugin layout: plugins/skein/ (Claude mirror) and plugins/skein-codex/ (Codex mirror) with intentional dispatch-idiom and path-anchor divergence (${CLAUDE_PLUGIN_ROOT} template substitution vs $SKILL_DIR env-export).
• Opt-in --auto-fix=trivial tier for /deep-review and /review-plan, gated by a single-source allowlist (scripts/auto-fix-allowlist.json) and bundled appliers (regenerated by just bundle-appliers).
• Repo invariants: canonical-↔-bundled byte-identity check, Claude-↔-Codex SKILL.md prompt parity check, trunk-snippet parity check, no-manual-apply-fallback enforcement.
• scripts/delete-skills.sh to surgically remove pre-plugin flat skill copies after a successful plugin install.
• Public GitHub install: both harnesses install directly from vr000m/skein without a local clone; local-clone form retained as a contributor workflow.

Changed

• Marketplace renamed from skein-local to skein in both .claude-plugin/marketplace.json and .agents/plugins/marketplace.json. Install command is /plugin install skein@skein (Claude) and codex plugin add skein@skein (Codex).

See CHANGELOG.md and README.md for full details.