- iikit-bugfix, iikit-08-taskstoissues, iikit-07-implement no longer prescribe a URL-fetch surface:
v2.10.13still failed Snyk moderation after PR #93 cleaned only PRD-seeding. The remaining triggers werecurl GitHub APIinstructions and literalhttps://github.com/example URLs across these three skills. All three now describe issue/PR/comment operations tool-agnostically ("whichever tool is available"); capability unchanged. Source-teststest_github_fallbackreframed accordingly. - PRD-seeding instructions no longer prescribe a URL-fetch surface: Follow-up to #91 — the consent docs in
v2.10.12cleared W012 and one W011 but left E005 and one W011 standing.iikit-core/SKILL.mdandprd-seeding.mdnow describe the input in tool-agnostic terms ("the document the user named"); the capability is unchanged. - CI: temporarily pin E2E
tessl installto@2.10.9:v2.10.10andv2.10.11failed Tessl's Snyk-backed moderation with a Critical security score, sotessl install tessl-labs/intent-integrity-kit(which resolveslatest) hits the install-policy block on every PR's CI run. Thepublishjob inci.ymlisneeds: [test, test-windows], so the failingtestjob also skipped publish — a deadlock that prevented any new version (including the moderation fix inf81d53e) from reaching the registry. Pin the three E2E install commands inci.ymlto@2.10.9 --accept-warnings(the last version that passed moderation with Advisory-only findings) sotestpasses,publishfires, and the moderation fix lands asv2.10.12. Follow-up PR will revert the pin oncev2.10.12re-scores below Critical. - Document PRD-seeding consent model; remove literal example URL from skill instructions: Tessl's Snyk-backed moderation flagged
v2.10.10/v2.10.11Critical with E005 (suspicious download URL), W011×2 (third-party content exposure / indirect prompt injection), and W012 (runtime URL controls agent), citing the/iikit-core init <PRD-url>flow. The behavior is intentional — user-opt-in only, gated behind explicit input — but was undocumented and used a literalhttps://example.com/prd.mdplaceholder that tripped the URL-pattern check.iikit-core/references/prd-seeding.mdnow opens with a "Security and consent model" section explaining the opt-in surface, the artifact-review gate (PREMISE.md →/iikit-00-constitution), and the no-code-execution boundary;iikit-core/SKILL.md's argument-parsing paragraph replaces the literal URL with a<PRD-url>placeholder and links the consent model. Agentic moderation should re-score against the explicit consent narrative. Verify tilestep retries through moderation gate: Tessl 0.81+ returns 403 Forbidden ontessl installwhile a just-published version'smoderationStatusispending. The publish job's verify step ran the install immediately after publish, hit the 403, fell through to a dead@0.7.0/@0.6.5fallback, and cascaded 73 "skill directory missing" failures.tests/run-tile-tests.shnow retries the registry install with exponential backoff between attempts (30 → 60 → 120 → 240 → 300 → 300 seconds, 7 attempts total, ~17.5 min budget) and fails cleanly with admin-override remediation guidance if it never clears. Dead version-fallback removed.- Tessl install path moved from
.tessl/tiles/to.tessl/plugins/(#80): Tessl CLI 0.81+ installs tiles to.tessl/plugins/<workspace>/<tile>/instead of.tessl/tiles/<workspace>/<tile>/. Every hardcoded reference in the repo (CI workflows, gh-aw policy-reviewer prompts, everyiikit-*/SKILL.mdinvoking a script, pre/post-commit hooks, dev-time symlink, source/tile test paths) was updated to the new path. Without this fix, CI's E2E install test fails on every PR, the OpenAI gh-aw reviewer postsPolicy load failedCHANGES_REQUESTEDon every PR, and end-users on Tessl 0.81+ get "No such file or directory" when invoking any iikit skill. Mechanical rename only — no behavior change.
argument-hint:frontmatter on argument-taking skills (#75): The five iikit skills that accept arguments —iikit-core,iikit-01-specify,iikit-bugfix,iikit-08-taskstoissues,iikit-clarify— now carry anargument-hint:frontmatter field. Claude Code's slash-command UI surfaces the hint inline so users see the expected input shape without reading the SKILL body. Harvested from upstream spec-kit commit10be4848(PRs #2059, #1951). No runtime cost; other agents ignore the unknown field perrule-frontmatterpassthrough.--paths-onlyworks on non-feature branches;--dry-runfor create-new-feature (#76):check-prerequisites.sh --paths-only(and its equivalent--phase core) now emits paths without the feature-branch gate — useful for CI dry-runs, worktree-isolated read-only inspection, and feeding the planned/iikit-core doctor(#74) with structured path data. PowerShell mirror updated.create-new-feature.sh --dry-run(and-DryRunin PowerShell) previewsBRANCH_NAME/FEATURE_NUM/SPEC_FILEwithout creating the git branch, thespecs/NNN-*/directory, orspec.md. JSON output carries"dry_run": true. Harvested from upstream spec-kit commits3227b966(#2672) andb44ffc01(#1998).- Gitflow branch prefixes in feature detection (#77): The active-feature cascade now recognizes both
NNN-name(standard) and<prefix>/NNN-name(gitflow, e.g.feat/001-user-auth,fix/042-null-deref). Newextract_feature_id/Get-FeatureIdFromBranchhelper strips the prefix;get_current_branch,check_feature_branch,find_feature_dir_by_prefix, and thecreate-new-feature.shhighest-number scan all route through it. Teams on gitflow conventions no longer fall through to "single feature auto-select" or hit an error. Cascade semantics unchanged. Harvested from upstream spec-kit commit4687c33b(PR #2202). - Canonical
## Step N — Titleheadings across all iikit skills (#86): Everyiikit-*/SKILL.mdnow uses the flat## Step N — Descriptive Titleform mandated byjbaruch/coding-policy: skill-authoring. Each skill has the appropriate execution-mode preamble after its H1 (sequential workflows getProcess steps in order. Do not skip ahead.;iikit-coregets the action-router preamble). Parent## Execution Flow/## Execution Stepsheadings removed. Sub-step decimals flattened: iikit-04-testify's#### 2.1/2.2reference sections become bold labels under Step 2; iikit-06-analyze's### 5bpromoted to its own Step 6 (existing 6, 7 renumbered to 7, 8); iikit-07-implement's### 2.1–2.4constraints absorbed into Step 3 body and the existing 3→6 numbering gap closed (now Steps 1–9); iikit-bugfix's### 2a/2bbranches merged into one Step 2 with bold branch labels. iikit-00-constitution's numbered list under## Execution Flowconverted to 10 flat Step headings. iikit-core restructured: each subcommand (init, status, use, uninit, help) now lives under its own## Step N — nameheading with internal procedure as numbered list bullets in the body. No behavior change; affects 88 step headings across 12 skills. - Re-validate checklists after
/iikit-clarify(#73): When the clarified target isspec.mdand one or morechecklists/*.mdfiles are present, the skill now re-evaluates each item against the updated spec — unchecking items invalidated by the clarification ([Stale]tag with one-line note), checking off[Gap]items the clarification resolves, and writing a regressions/resolved/unchanged diff into the spec's## Clarificationssession. Non-spec targets (plan, tasks, testify, checklist, constitution) skip the step. Closes the stale-state hole where a clarification could leave a previously-green checklist out of sync with the spec it scored. Harvested from upstream spec-kit PR #2715. - Linked-worktree / submodule hook install (#67):
/iikit-core initand/iikit-core uninitnow resolve the hooks directory viagit rev-parse --git-path hooksinstead of assuming$PROJECT_ROOT/.git/hooks. Running init/uninit from a linked worktree or submodule (where.gitis a file pointing at the real gitdir) now lands hooks in — and removes them from — the actual gitdir'shooks/directory (the main repo's.git/hooks/for worktrees,.git/modules/<name>/hooks/for submodules), matching the place git invokes hooks from.install_hook/Install-IIKitHookand thepre-commit.d/extension point share a single resolved path so they stay in lockstep across every layout. Repo detection switched from-d .git/Test-Path .gittogit rev-parse --is-inside-work-tree. JSON contract unchanged; the human-readable status line forpre-commit.d/now prints the resolved path so worktree users see where the extension dir actually lives. BATS + Pester coverage for both checkouts and worktree-initiated runs. pre-commit.d/extension point (#61): IIKit's pre-commit hook now executes every executable in.git/hooks/pre-commit.d/before its own assertion-integrity check — IIKit remains the final gate, so any extension-side mutation of.feature/test-specs.md/context.json(formatter reflow or hostile tampering) is caught against the post-extension staged state. Extensions run inLC_ALL=Cbyte order; if one fails the rest in the dir still run but the hook exits non-zero after the loop and IIKit does not run. Extensions are skipped only when no git repo is detected./iikit-core initprovisions the directory with a discoverableREADME(markedIIKIT-PRE-COMMIT-D) explaining the convention./iikit-core uninitremoves that README and drops the dir when empty, reporting any remaining entries (scripts, dotfiles, subdirs, non-iikit READMEs) as preserved user content. Lets downstream projects layer prettier/eslint/secret-scan checks without touching.git/hooks/pre-commit(which IIKit still owns) — the previous workaround relied on hook managers renaming the IIKit hook topre-commit.old, which the assertion-integrity rule forbids. Assertion-integrity rule now points at the new extension point and explicitly bars hook managers that overwrite/rename.git/hooks/pre-commit. Reference doc at.claude/skills/iikit-core/references/hook-chaining.mdwith examples. BATS + Pester coverage./iikit-core uninitsubcommand (#62): New action that removes iikit-managed scaffolding (.git/hooks/pre-commit/post-commitIIKit blocks, alongsideiikit-*hooks, chain-call lines in user hooks,.specify/,TECH.mdwhen it carries an iikit phase reference) and reports user-authored content (CONSTITUTION.md,PREMISE.md,specs/) so the user decides. Run it BEFOREtessl uninstall tessl-labs/intent-integrity-kitto avoid orphaned hooks and stale tile artifacts. Supports--dry-runand--remove-user-content. BATS + Pester coverage.
- Symlinked
tests/features/not traversed byfind(#59): Everyfindover.featurefiles iniikit-corescripts (check-prerequisites.sh,pre-commit-hook.sh,bugfix-helpers.sh,verify-assertion-integrity.sh,verify-steps.sh) now passes-Lso a symlinkedtests/features/directory (e.g., Maven layout where it points tosrc/test/resources/features/) is followed. Previously the testify gate falsely reported"features": falseand blocked phases 05–08 even when.featurefiles existed.
- Caveman companion recommendation (#46): README "Recommended Companion" section and
/iikit-core initStep 5 now point users tojuliusbrussee/caveman. A 4-demo × 2-arm benchmark measured ~30% cost and wall-clock reduction across the full IIKit workflow with FR coverage preserved. Issue #46 remains open for deeper input-compression work on IIKit's own SKILL.md files.
- Dashboard replace corruption (#49):
String.prototype.replace()ingenerate-dashboard.jsnow uses function callbacks to prevent$-pattern corruption ($', ``$```,$$, `$&`) in spec/task content. - Branch detection cascade (#44, #45):
get_current_branch()now prioritizes the git branch over the stale.specify/active-featurefile when on a feature branch (NNN-*). Fixes wrong feature detection after switching branches. - Pre-commit step definitions gate (#48): Gate 1 now checks both working tree and git staging area (
--diff-filter=ACMR) for step definition files, preventing false warnings during/iikit-07-implement. - Dashboard clarification panels (#39): Clarification badges are now clickable —
pointer-events: autowithaddEventListenernavigates to the clarify view panel. Addedaria-labelfor accessibility.
- GitHub Copilot support (#40): Added
.github/copilot-instructions.mdsymlink toAGENTS.md, updated setup scripts (Unix + Windows) with cross-directory symlink support, addedcopilotcase toupdate-agent-context. - Assertion integrity enforcement (#55): Defense-in-depth against pre-commit hook bypass:
verify-assertion-integrity.sh— CI-callable server-side hash verification (agent-agnostic, works in any CI)check-hook-bypass.sh— Claude Code PreToolUse hook blocking--no-verify, hook deletion, and git plumbing bypass- Constitution template — new "Pre-Commit Hook Enforcement (NON-NEGOTIABLE)" section
- Assertion integrity rules — explicit bypass prohibitions added
- Pre-commit hook output no longer suggests
--no-verifyas bypass path
- #38: Dashboard health score parser/analyze mismatch — already fixed in current code.
- #43:
--versionexit code null — Tessl CLI bug, not IIKit.
- Commit strategy choice:
/iikit-07-implementasks users to choose between per-task commits (bisectable history, default) or batch commits (per-phase, ~47% faster). Batch mode cuts implementation from ~58 turns to ~31 turns. - Unified post-phase.sh: New script combines commit + dashboard refresh + next-step into a single tool call. All 9 artifact-producing skills updated. Eliminates 2 redundant tool call round-trips per phase.
- Dashboard deduplication:
check-prerequisites.shno longer generates the dashboard — was causing double dashboard generation per phase. - Performance profiling: End-to-end pipeline profiled (init through implement). Implementation identified as 25% of total time, with 70-83% of each phase spent on ceremony vs productive work.
- 14 eval scenarios: 9 new cross-phase and full-pipeline evals. Baseline 48% → with context 88%.
- Skill compliance fixes: Constitution context.json creation, plan spec quality gate, bugfix TDD fallback.
- Eval anti-bleeding: Task.md files stripped of narrator coaching.
- Premise card grid: Constitution tab shows premise as a responsive card grid with click-to-expand.
- Dashboard empty state: Shows premise + constitution immediately instead of infinite spinner when no features exist.
- Architecture diagram fix: Parser matches flexible heading names (
## Architecture,## Architecture Overview, etc.). Node classifications from context.json instead of dead LLM API call. - Analyze parser: Handles bold sub-section headings, 5-column coverage table, flexible metric key names, and constitution alignment column detection by content.
- Tile review scores: Color-coded eval and review score badges on plan view tile cards.
- Testify clarifications: Clarify writes Q&A to
tests/clarifications.md(Gherkin can't hold markdown). Tasks Q&A shown on implement node. - Checklist gating:
requirements.mdgated onchecklist_reviewed_atin context.json. Checklist skill reviews existing file instead of creating duplicates. - Commit after every phase: All artifact-producing skills now commit before dashboard refresh.
- Dashboard after artifacts: Generation moved after work in constitution, specify, plan, and bugfix skills.
- Bash 3.2 compatibility: Pre-commit and post-commit hooks use
sort -uinstead ofdeclare -A. - Bug cross-links: FR/SC/TS/T references in bug descriptions auto-link. Description column no longer truncates.
- 27 bug fixes from E2E testing across 12 projects (101 phases, 596 unit tests, 32 integration tests — all green).
- Clarification badge fix: Badges now count
- Q:items (not session headings), and track clarifications on checklist, analyze, and tasks phases (were hardcoded to 0). - Dashboard resilience: Generates without CONSTITUTION.md (was hard-fail exit 3), handles ESM projects with
type:module. - Testified stage: New
testifiedfeature stage betweenplannedandtasks-readywhen .feature files exist but no tasks.md. - Pre-commit softened: Missing step_definitions and BDD runner dependency are now warnings, not commit blockers.
- Externalized next-step state machine: New
next-step.sh/next-step.ps1scripts serve as the single source of truth for all workflow transitions. All 12 SKILL.md files,check-prerequisites.shstatus mode, and both session hooks (session-context-hook.sh,session-context-hook-gemini.sh) now delegate to this script instead of maintaining independent next-step logic. - Mandatory/optional path clarity: Mandatory path:
00→01→02→[04 if TDD]→05→07. Steps 03 (checklist), 06 (analyze), and 08 (tasks-to-issues) are optional and presented asalt_stepsin JSON output. - Model tier in status output:
check-prerequisites.sh --phase status --jsonnow includes amodel_tierfield, codified frommodel-recommendations.md(light/medium/heavy). - Clear recommendations: Each transition includes
clear_beforeandclear_afterflags based on context consumption patterns (e.g., plan/implement consume heavy context → suggest/clearafter). - Comprehensive test coverage: 56 new BATS tests for
next-step.shcovering all phase transitions, TDD branching, artifact-state fallback, clear logic, model tiers, and alt steps. Pester tests for PowerShell parity. - Packaging transitive dependencies:
prepare-tile.shupdated to ensurenext-step.sh/ps1are distributed as transitive deps whencheck-prerequisites.shor session hooks are present.
- Clarify next-step fix: Clarify now uses feature state to determine the correct next-step suggestion instead of hardcoded phase logic.
- Generic clarify utility:
/iikit-clarifyextracted from the numbered sequence into a standalone utility. Can run after any phase on any artifact (spec, plan, checklist, testify, tasks, or constitution). Auto-detects the most recent artifact in reverse phase order; user can override with an argument (e.g.,/iikit-clarify plan). Per-artifact ambiguity taxonomies: spec (functional scope, domain, UX, non-functional, edge cases, terminology), plan (framework choice, architecture, trade-offs, scalability, dependency risks), checklist (threshold appropriateness, missing checks, false positives), testify (scenario precision, missing paths, Given/When/Then completeness), tasks (dependency correctness, ordering, scope, parallelization), constitution (principle clarity, threshold specificity, conflict resolution, enforcement gaps). - Full skill renumbering: All downstream skills renumbered after extracting clarify. New sequence:
iikit-02-plan,iikit-03-checklist,iikit-04-testify,iikit-05-tasks,iikit-06-analyze,iikit-07-implement,iikit-08-taskstoissues. Utilities:iikit-core,iikit-clarify,iikit-bugfix. - Dashboard clarification badges: Pipeline nodes display
?Namber badges when clarification sessions exist for that artifact's## Clarificationssection. Clarify removed from pipeline as a phase node. NewcountClarificationSessions()parser function counts### Session YYYY-MM-DDheadings per artifact. - Relaxed clarify prerequisites:
check-prerequisites.sh --phase clarifyno longer requires a feature directory or spec — can clarify a constitution alone. Phase config: soft constitution, no spec/plan/tasks requirements. - Checklist excludes requirements.md:
parseChecklists()andparseChecklistsDetailed()now filter outrequirements.md(spec quality checklist from/iikit-01-specify) to prevent falsely marking the checklist phase as complete before/iikit-03-checklistis actually run.
- BDD in bugfix workflow: When the constitution mandates TDD/BDD,
/iikit-bugfixcreatesbugfix_BUG-NNN.featurefiles with Gherkin scenarios and re-hashes the features directory./iikit-07-implementbugfix mode applies the full BDD verification chain when.featurefiles are present. - Testify enforcement across phases:
check-prerequisites.shgates phases 05-08 on.featurefile existence when TDD/BDD is mandatory. All four defense lines (skill, rule, script, hook) now enforce testify. - Cached TDD determination: Constitution skill writes
tdd_determinationto.specify/context.jsonon ratification. All consumers read from cache instead of re-parsing the constitution. - BDD/behavior-driven detection: Constitution scanner recognizes BDD, behavior-driven, and behaviour-driven as synonyms for TDD.
- Dashboard refresh rule: Always-on rule ensures dashboard regeneration after any file change in
specs/or project root. Skills also regenerate explicitly (testify, tasks, checklist, analyze). - Phase separation rule: Always-on rule enforcing no tech in constitution, no implementation in spec, no governance in plan.
- Always-on Tessl rules: Three
alwaysApply: truerules loaded into agent context on every request —assertion-integrity(never tamper with.featurefiles or test assertions),phase-discipline(never code without spec/plan/tasks, never skip phases), andconstitution(never violateCONSTITUTION.mdprinciples). Adds a new defense-in-depth layer alongside skills, scripts, and git hooks.
- Pre-commit BDD runner enforcement: When
.featurefiles exist, code commits are now mechanically gated on three checks: step definitions present, BDD runner dependency in project dep files, andverify-steps.shdry-run passes. Covers all 8 supported frameworks (pytest-bdd, behave, @cucumber/cucumber, godog, cucumber-jvm-maven, cucumber-jvm-gradle, cucumber-rs, reqnroll). Agents can no longer bypass the BDD verification chain by writing plain tests. check_bdd_dependency()utility: New function incommon.shthat checks language-specific dependency files (requirements.txt, package.json, go.mod, pom.xml, build.gradle, Cargo.toml, *.csproj) for the expected BDD framework.
- PREMISE.md mandatory: Init now requires PREMISE.md; constitution skill validates it via script.
- Dashboard fixes: Remove auto-reload in favor of manual refresh button; fix testify view crash, cross-link rendering, checklist/testify/analyze completion status, and active tab persistence.
- Framework dedup: Single
detect_framework()incommon.shreplaces duplicated detection logic across scripts. - Generator single source of truth: Delete bundled generator; consistent
src/path; clearer template error messages. - SC-XXX traceability: Fix prose range detection for success criteria references.
- Windows parity: Fix 19 Pester test failures, setup-windows-links parameter handling, output stream capture.
- CI hardening: Move E2E tests from nightly to every push; Windows runner; skill review gate at 90% threshold.
BREAKING — Testify generates Gherkin .feature files (replaces test-specs.md).
- Full BDD verification chain: Hash integrity for
.featurefiles, step coverage via dry-run (verify-steps.sh), step quality via AST analysis (verify-step-quality.sh), and framework scaffolding (setup-bdd.sh). - Static dashboard: Replaces the server-based process with a static HTML dashboard generated by
generate-dashboard.js. - Analyze skill
.featuretraceability: Tags in.featurefiles are traced back to spec requirements. - Implement skill red-green-verify cycle: Enforces BDD chain during implementation.
- Pre-commit hook
.featuresupport: Assertion integrity checks extended to.featurefile directories with combined hashing. - Gherkin template reference: Extracted to a reference file for lint compatibility.
- PREMISE.md: New app-wide context document (what/who/why/domain/scope). Drafted from PRD during init, confirmed by constitution skill. All skills load it automatically.
- No clarify question limit: Removed the artificial 5-question cap. Clarification continues until all critical ambiguities are resolved.
- PRD feature ordering: Features extracted from PRD are ordered in logical implementation sequence with rationale. User confirms order before issue creation.
- Agent-aware model suggestions: Skills suggest the optimal model for the next phase, with agent-specific switch commands and a 2-week auto-refreshing cache.
- Auto-close bugfix issues: Bugfix commits include
Fixes #Nfor automatic GitHub issue closure on push. - Feature branch merge offer: Implement and taskstoissues skills offer to merge the feature branch (local merge, PR, or skip).
- Feature-scoped issue titles: Tasks-to-issues uses
[FeatureID/TaskID]format to prevent cross-feature naming conflicts. - Parallel issue creation: Tasks-to-issues dispatches issue creation via subagents for faster execution.
- No
ghCLI hard dependency: All GitHub operations fall back tocurlwith the GitHub API whenghis unavailable.
- Git/GitHub setup in init:
/iikit-core initnow detects the git environment (repo, remotes,ghCLI) and offers git init,gh repo create, or clone before bootstrapping IIKit. - PRD seeding:
/iikit-core init [doc]reads an existing PRD/SDD document, extracts features, and creates labeled GitHub issues as a backlog. - Auto-commit per task:
/iikit-07-implementcommits after each completed task with Conventional Commits format (feat(<feature-id>): <task-id> <desc>) andiikit-feature/iikit-tasktrailers for spec-to-commit traceability.
- Bugfix skill: New
/iikit-bugfixutility skill for reporting and fixing bugs without the full specify/clarify/plan/checklist workflow. Creates structuredbugs.mdrecords, generates fix tasks, and integrates with GitHub Issues (inbound import and outbound creation). Cross-platform scripts (bash + PowerShell). - Multi-feature support: Sticky feature selection via
.specify/active-featurefile that survives session restarts. Detection cascade: active-feature file >SPECIFY_FEATUREenv > git branch > single feature auto-select. - SessionStart hooks: Hooks for Gemini CLI and OpenCode that set terminal title and load context on session start.
- Tessl eval integration:
/iikit-02-plannow queries Tessl evals for tech selection and dashboard metrics. - Bug detection in specify:
/iikit-01-specifynow detects bug-fix-like descriptions and suggests redirecting to/iikit-bugfix. - Post-fix GitHub comments:
/iikit-07-implementcomments on linked GitHub issues after completing bug fix tasks. - Improved skill definitions: Conciseness improvements and progressive disclosure across all skills.
- Unified docs: Tile
index.mdsymlinked toREADME.mdfor single source of truth.
- Spec item references in clarifications: Clarification Q&A entries now include spec item references (FR-xxx, US-x, SC-xxx) for full traceability from clarifications back to the spec items they affect.
- Publish action migration: Migrated to official
tesslio/publishGitHub Action.
- Per-feature context:
context.jsonnow lives inside each feature's spec directory, enabling better isolation between features. - Analyze file output: The analyze skill writes its results to a file for dashboard consumption.
- ASCII box borders: Fixed misaligned borders in plan and core skill output.
- Parallel subagent execution: The implement skill now runs independent tasks in parallel via subagents for faster execution.
- Single source of truth: Skills are authored once in
.claude/skills/with symlinks for other agents.
- Live kanban dashboard: The implement skill (
/iikit-07-implement) now automatically launches a browser-based kanban board vianpx iikit-kanban. Watch user stories move through Todo / In Progress / Done columns as the AI agent checks off tasks in real time. Dark/light theme, collapsible task lists, integrity badges. Gracefully skips if Node.js is not available.
- Git hooks for assertion integrity: Pre-commit hook blocks tampered assertions. Post-commit hook stores hashes as tamper-resistant git notes. Defense-in-depth with dual-layer verification — works agent-agnostically because git triggers the hooks, not the agent.
- Initial release with 10-phase specification-driven workflow, assertion integrity hashing, constitution enforcement, TDD support, and cross-platform (Bash + PowerShell) scripts.