release-notes.md

Release Notes

Version 0.9.0

Added

• Interactive tutorial: cicadas init now prompts first-time users to run a 7-step interactive tutorial. The tutorial walks through the full Cicadas workflow with mock output, concept explanations, and agent prompts at each step. Run any time via 💬 "Run the Cicadas tutorial" or python src/cicadas/scripts/cicadas.py tutorial. Supports --tutorial (auto-run) and --no-tutorial (skip prompt) flags on init.
• Next-step hints: Every lifecycle command (kickoff, branch, archive, update_index, open_pr, init) now prints a 💬 next-step hint after its main output. cicadas status infers and prints a hint from registry state (no .cicadas/ → init; no initiatives → start; no branches → implement; no lifecycle → complete partition).
• Hint subsystem: utils.py provides hints_enabled(args, config) and print_hint(lines, args, config) (66-char cyan box, auto-suppressed when not a TTY). Disable globally via hints: false in .cicadas/config.json, or per-command with --no-hints.

Changed

• README workflow section: Rewrote the Getting Started workflow as a 7-step narrative using 💬 agent prompts. No raw CLI commands exposed to users. Added Quick Reference cheatsheet table.
• HOW-TO.md: Added Interactive Tutorial section (how to run, step overview) and Hint Toggling section (global config, per-command flag, CI auto-suppression).
• Default branch convention: Greenfield Cicadas repositories and test fixtures now default to main, while existing master repositories and remote HEAD configuration remain supported.

Version 0.8.2

Deprecated

• Skill builder removal: Removed the built-in skill-builder triggers from src/cicadas/SKILL.md and marked the legacy skill-authoring docs as deprecated.

Version 0.8.1

Added

• Initiative profiles: The standard start flow now records product, technical, or mixed initiative profiles. Technical initiatives can use a new Technical Brief template and optional Operator Experience template while Tech Design, Approach, and Tasks remain mandatory for cross-module work.
• Template coverage: tests/test_templates.py now verifies the technical-profile templates share the compact front matter contract and that profile-aware guidance stays documented.

Fixed

• Review finding fixes: Addressed the code review findings for initialization safety, deterministic git-backed tests, code graph extraction, token log concurrency, active spec path naming, safe synthesis apply behavior, and module context path containment.
• Init preservation: Re-running init now preserves existing .cicadas JSON state instead of overwriting project registry, index, config, or active work.
• Synthesis path safety: synthesize --apply now rejects canon update paths that resolve outside .cicadas/canon/.
• Token log concurrency: tokens.py now serializes append operations with a lock and atomic replacement so concurrent writers do not lose entries or corrupt tokens.json.
• Lightweight active specs: Archive, prune, and synthesis helpers now resolve active directories consistently for lightweight fix/, tweak/, and skill/ branches, while preserving fallback compatibility for legacy names.
• Graph edge preservation: Streaming Python and Java graph extraction now preserves linked-test edges through staged writes and keeps Java test node IDs consistent.
• Regression coverage: Added focused tests for the above edge cases, including concurrent token appends, path traversal rejection, graph streaming edge preservation, deterministic temp git defaults, and out-of-repo module context rejection.

Changed

• UXD mock-up contract: Screen-based UX drafts now require at least one editable HTML/CSS mock-up under .cicadas/drafts/{initiative}/mockups/ before UX approval. The ux.md template includes a dedicated mock-up section, screenshot previews remain optional, and operator-only flows still use operator-experience.md.
• Scan exclusions: scan_repository now skips directories containing a SKILL.md file (SDD installs and agent skill bundles at arbitrary paths), root-level hidden/underscore dirs whose names contain a known SDD tool substring (bmad, cicadas, gsd, openspec), and any paths listed in the new scan_exclude_paths key in .cicadas/config.json. This prevents SDD state/output directories like _bmad-output/ and vendored SDD installs from inflating meaningful_file_count, estimated_loc, and slice candidates.

Version 0.8.0

Added

• Adaptive canon bootstrap that classifies repos as normal-repo, large-repo, or mega-repo, writes repo.json / repo-tree.jsonl / repo-context.md, and seeds lazy slices/ packs only for larger repos.
• Code Graph DB (currently supports python and java) to assist the coding agent in navigating large- and mega-repos.

Changed

• Initiative completion now uses targeted canon reconcile for large and mega repos, updating touched slices by default and refreshing top-level orientation docs only when durable repo-wide truth changed.
• init no longer creates an empty .cicadas/canon/modules/ directory up front.
• Optional code graph builds now write progressively to staged SQLite tables under .cicadas/graph/, emit progress.json / progress-log.jsonl with elapsed time and ETA, and expose graph tail / graph watch for long-running local builds.
• Code graph routing is more useful on large repos: deterministic area-plan.json output now captures seeded routing areas, graph search adds graph-native file/symbol/entrypoint search, --exclude-tests filters operational hunts, and JavaScript/TypeScript now have structural graph indexing.
• Java graph extraction now uses a structural baseline plus resumable semantic enrichment batches. Successful semantic batches are retained even when some files fail, bad files are quarantined via batch bisection, and graph metadata now reports structural, semantic, or hybrid Java coverage.
• Repo scan now excludes local Cicadas workspaces like .cicadas-skill/ from inventory, scale heuristics, and graph routing seeds so repo graphs stay focused on operational code.

Version 0.7.5

• Context Optimization: The five core initiative templates (prd.md, ux.md, tech-design.md, approach.md, tasks.md) now share a compact machine-readable front matter contract with summary, modules, depends_on, and index fields so agents can reload approved context without re-reading full drafting threads.
• Reset Rules: SKILL.md now defines explicit Branch Reset, Phase Reset, and Partition Reset rules. These rules re-anchor agents on approved file-backed state, reload indexed sections first, and opportunistically ask the host to clear or compact context when supported.
• Clarify / Canon Routing: emergence/clarify.md now refreshes the new front matter fields instead of steps_completed, templates/canon-summary.md includes a branch-start routing cue for compact context loading, and the synthesized canon now documents the compact-context workflow.
• Verification: Added tests/test_templates.py to verify the front matter contract, Clarify front matter guidance, and canon summary routing hint.
• Common CLI Fix: cicadas.py now forwards option-style subcommand arguments like update-index --branch ... --summary ... correctly instead of rejecting or misrouting them, and tests/test_cli.py includes a regression test for wrapper-based update-index.
• Worktree Ergonomics: Initiative and lightweight (fix/, tweak/, skill/) worktrees are no longer created implicitly by default. .cicadas/config.json now supports auto_worktrees.initiatives, auto_worktrees.lightweight, and auto_worktrees.parallel_features, and both kickoff.py and branch.py accept --worktree for explicit opt-in.
• Safer Parent Resolution: branch.py now resolves parent refs against both local branches and origin/..., so feature creation still works when the intended initiative parent exists only as a remote-tracking branch.
• Visibility & Coverage: status.py now lists initiative worktrees, check.py warns on stale initiative worktrees, and new integration tests cover config-enabled worktrees, remote-only parent refs, and worktree creation failure handling.
• Common CLI Surface: Added src/cicadas/scripts/cicadas.py and command_registry.py as the repo-local command entrypoint for deterministic Cicadas operations. The new surface wraps lifecycle, review, event, skill, and token commands behind one agent-discoverable --help tree while preserving the underlying scripts for compatibility.
• CLI Coverage: Added tests/test_cli.py to verify top-level help, representative command dispatch, lifecycle mutation through the wrapper, token subcommands, update-index forwarding, and clean failure on malformed tokens show --full input.
• Docs: Updated the canonical docs to explain the compact context contract, reset-boundary behavior, common CLI usage, and the new worktree defaults. Evals remain future work; these releases only add the scaffolding for better context hygiene.

Version 0.7.0

• Machine-Testable Partition Specs: approach.md template now includes #### Artifact Type, #### How to Run, and #### Acceptance Criteria subsections in each partition block. emergence/approach.md (Step 4b) guides agents to infer artifact type, generate How to Run from package.json/pyproject.toml/Makefile/Dockerfile, and produce acceptance criteria by artifact type (rest-api → endpoint assertions, web-ui → interaction criteria, cli → stdout/exit, library → return values, background-service → side effects, full-stack → combined). <!-- NEEDS MANUAL REVIEW --> flags non-deterministic or external-dependency criteria.
• Event Log Infrastructure: New emit_event.py and get_events.py scripts. Each initiative maintains an append-only events.jsonl at .cicadas/active/{initiative}/events.jsonl. emit_event.py writes typed events with fcntl.flock for concurrent-write safety. get_events.py reads and filters by --type (exact or prefix), --since (ISO 8601), and --last N; outputs JSONL to stdout; exits 0 with empty output when events.jsonl is absent.
• Event Log Integration: Lifecycle scripts now emit events automatically: kickoff.py → initiative.kicked_off; branch.py → branch.created and worktree.created; archive.py → specs.archived; open_pr.py → pr.opened (gh/glab success) and pr.blocked (BLOCK verdict). All calls use utils.emit() — a non-fatal lazy-import wrapper so emit failures never abort primary operations.
• Status event display: status.py now shows the 5 most recent events per active initiative in the status output, with graceful handling of missing events.jsonl.
• Implementation Rules 9 & 10: implementation.md adds Rule 9 (emit task.complete after each task checkbox) and Rule 10 (emit partition.complete before partition PR, with summary, canon_entry, and notes_for_evaluator fields).
• SKILL.md Event Log section: New "Event Log" subsection in Operations documents the event log path, emit_event/get_events CLI, event schema, full lifecycle and agent event type tables, and the design invariant (get_events.py is the only consumer interface — direct file reads are forbidden).
• utils.emit(): Shared non-fatal event emitter added to utils.py. Replaces per-script _emit helpers; all lifecycle scripts import emit from utils.
• Tests: 16 new tests for emit_event.py and get_events.py (concurrent writes, type prefix matching, --last, --since, missing file); 4 new integration tests asserting events land in events.jsonl after kickoff, create_branch, and archive; 1 new test asserting status.py displays Recent events. 251 tests total.

Version 0.6.3

• Worktree Support: kickoff.py now creates a linked git worktree (../{repo}-initiative-{name}) instead of switching the main worktree to the new branch. branch.py extended to also create worktrees for fix/, tweak/, and skill/ branches (previously only parallel feat/ partitions used worktrees). Multiple Cicadas workstreams can now run simultaneously without branch-stomping.
• Registry Root Detection: New get_registry_root() and get_registry_dir() in utils.py. Detects linked worktrees via .git file vs directory and routes registry.json/index.json I/O to the primary worktree, keeping shared state authoritative across all linked worktrees.
• archive.py: Added initiative-level worktree removal on archive (branch worktrees already worked; initiatives now cleaned up too).
• Tests: CicadasTest.tearDown now cleans up linked worktrees before deleting temp repos. New tests for get_registry_root(), get_registry_dir(), and kickoff worktree behavior. Updated branch and orchestration tests to reflect new worktree-first behavior (main worktree stays on default branch).

Version 0.6.2

• 1-PR Archive Flow: archive.py now snapshots the initiative's registry metadata into a .cicadas_metadata.json file within the archive folder. This allows you to include the archive move and registry cleanup in your main PR for a single-commit finalization.
• Unarchive Script: New unarchive.py tool. If a PR needs rework after being archived, run cicadas unarchive {name} to instantly restore the registry state and move files back to active/.
• Integrated Guidance: open_pr.py now detects initiative branches and recommends the 1-PR flow. lifecycle-default.json updated to prioritize Synthesize → Archive → Open PR sequence.
• Tests: New tests/test_unarchive.py validates metadata snapshotting and state restoration.

Version 0.6.1

• Agent Skill Authoring: New skill/ branch prefix and full skill lifecycle. skill-create.md — dialogue-driven authoring: start flow → 4-question clarifying dialogue → complete SKILL.md + optional bundled files (scripts/, references/, assets/) + eval_queries.json draft → kickoff + branch + validate. skill-edit.md — one diagnostic question, minimum-change before/after proposal, validate on apply. validate_skill.py — spec-compliance validator (name charset/length/dir-match, description ≤1024 chars, frontmatter delimiters; stdlib regex, handles single-line, folded, and block YAML scalars). skill_publish.py — copy or symlink active skill to publish_dir with pre-publish validation gate; reads destination from emergence-config.json. skill-SKILL.md scaffold template. skill/ added to branch.py, abort.py, archive.py, status.py (new "Active Skills" display section). start-flow.md updated: new Publish destination step (skills only), skill column in scoping table, Building on AI? skips eval-status follow-up for skills. SKILL.md updated: branch hierarchy, Skills Operations section, Builder commands, CLI entries. 22 new tests (14 for validate_skill, 8 for skill_publish) plus skill/ regression assertions across 4 existing test files. 174 tests total.

Version 0.6.0

• Building on AI: Start flow now asks "Is this project building on AI? (yes/no)" and, if yes, eval status (already have / will do). Choices are stored in emergence-config.json. Initiatives with "will do" evals: optional eval-spec authoring (template + LLMOps playbook) after PRD/UX/Tech; Approach asks eval placement (before build / in parallel) with parallel warning. Tweaks and bug fixes with "will do": optional eval/benchmark reminder in tweaklet/buglet. Cicadas does not run or host evals. New emergence/eval-spec.md instruction module and templates/eval-spec.md. SKILL.md, HOW-TO.md, CLAUDE.md, and canon updated.
• parse_partitions_dag fallback: When PyYAML is not installed, utils.parse_partitions_dag() uses a regex-based fallback to parse the yaml partitions block so tests and branch/worktree logic work in minimal environments.

Version 0.5.6

• Instruction module terminology: Replaced all uses of "subagent" with "instruction module" throughout SKILL.md, EMERGENCE.md, and all emergence/*.md files. Emergence files are inline role-switches in the current context window — no separate agent process is spawned. The old term was misleading to integrators.
• Prompt injection guardrails: Added explicit untrusted-input warnings to clarify.md (before reading requirements.md / loom.md) and bootstrap.md. Added corresponding Guardrail #8 to SKILL.md.
• Balanced elicitation inlined: Deleted emergence/balanced-elicitation.md (28-line file); content inlined as a full appendix in clarify.md. References in tech-design.md and ux.md updated to point to the appendix.
• Process banner consistency: Added FOLLOW THIS PROCESS EXACTLY. DO NOT SKIP STEPS UNLESS INSTRUCTED. to all 7 emergence files (clarify, ux, tech-design, approach, tasks, tweak, bug-fix). Was previously present on only 2, implying the others were optional.
• Mode selection decision rule: tasks.md now states when to use Foundation Mode vs. Feature Mode — greenfield/standalone → Foundation; existing system → Feature.
• Resuming Mid-Initiative: New ## Resuming Mid-Initiative section in SKILL.md guides context recovery when picking up an in-progress session.
• Branch cleanup: Added branch deletion steps to both "Complete an Initiative" (Step 4) and "Lightweight Paths" (Step 7) in SKILL.md.
• Bug fixes: Fixed approach.md asking for PR preference when lifecycle.json already exists (now skips); fixed ux.md Canon Check reading product-overview.md instead of ux-overview.md; fixed bug-fix.md and tweak.md using {initiative} instead of {name} in artifact paths.
• Implementation rules: implementation.md is now the single canonical source for implementation guardrails; SKILL.md contains a pointer only (eliminates drift risk).
• Script failure recovery: Added Guardrail #9 to SKILL.md with recovery guidance when a script fails mid-operation.

Version 0.5.5

• Standard start flow: When starting an initiative, tweak, or bug, the agent now runs a single mandatory sequence first: name (confirm even if given) → create draft folder → requirements source and pace (initiatives only) → PR preference → then collect requirements or draft the spec. New emergence/start-flow.md is the source of truth; Clarify, Tweak, and Bug Fix subagents embed this flow. EMERGENCE.md, SKILL.md, and HOW-TO.md updated; README, src/cicadas/README, agents.md, and CLAUDE.md reference the flow.
• Clarify intake options: At the start of Clarify, the Builder can choose how to provide requirements: Q&A (interactive), Doc (place a requirements document at .cicadas/drafts/{initiative}/requirements.md), or Loom (record in Loom, then save the transcript to .cicadas/drafts/{initiative}/loom.md). The agent fills the PRD from the doc or transcript. EMERGENCE.md now includes a "Requirements intake (Clarify)" subsection for discoverability. New regression test tests/test_emergence_clarify.py asserts that clarify.md documents the Intake step and file paths.

Version 0.5.4

• Context Injection at Branch Start: branch.py now writes context.md to the project root for sequential (non-worktree) branches, matching the existing behavior for parallel/worktree branches. Every registered branch now receives a canon summary + scoped module snapshots + approach.md + tasks.md at start. New templates/canon-summary.md template added. synthesis-prompt.md updated with Step 5 to produce canon/summary.md (300–500 token agent-optimized snapshot) as part of every canon synthesis run. SKILL.md synthesis step updated accordingly.

Version 0.5.3

• Code Review as Merge Gate: Code review is now a persistent artifact, not an ephemeral console report. The Code Review subagent writes a structured review.md to .cicadas/active/{initiative}/ with one of three verdicts: PASS, PASS WITH NOTES, or BLOCK. New scripts/review.py reads the verdict and returns exit codes (0=PASS/PASS WITH NOTES, 1=BLOCK, 2=not found). open_pr.py checks the verdict before opening a PR and refuses on BLOCK. Verdict strings updated throughout emergence/code-review.md (was MERGE-READY/NEEDS-WORK). New templates/review.md template. 13 new tests.

Version 0.5.2

• Token Usage Logging: New tokens.py module (init_log, append_entry, load_log) adds an append-only tokens.json log to every initiative. Created at kickoff, moves with specs through active → archive. Scripts write null phase-boundary entries; agents self-append real counts when the runtime exposes them. history.py now renders a per-initiative token summary (input/output/cached by phase) in the HTML timeline. 25 new tests.
• Emergence Pace Selection: clarify.md step 0 now asks the Builder their preferred review cadence — section (pause after each section), doc (pause after each complete doc, default), or all (draft everything then present). Choice is stored in emergence-config.json in the drafts folder; every subsequent emergence agent (UX, Tech Design, Approach, Tasks) reads it and enforces the stop rule.
• PR Boundary Reminder: clarify.md Process Preview now explicitly notes that Approach will ask about PR boundaries and create lifecycle.json, preventing agents from skipping that step.
• Implementation guardrails: Require Reflect and task updates before every commit on feat/task branches; require Code Review before committing on feature branches. implementation.md and SKILL.md updated; new "Implementation agent rules (all environments)" section in SKILL so Cursor and other non–Claude Code envs get the same guardrails from the skill file alone.
• Install & docs: Cursor integration now copies SKILL.md (was skill.md). README, HOW-TO, and CLAUDE.md clarify that CLAUDE.md is Claude Code–only and other envs use the skill for guardrails.
• Installer: Added rovodev agent integration — bash install.sh --agent rovodev creates .rovodev/skills/cicadas symlink to the install dir. Documented in README.md, HOW-TO.md, and src/cicadas/README.md.

Version 0.5.1

• Code Review: New optional agent operation (emergence/code-review.md) for spec-anchored evaluation of code diffs. Covers task completeness, acceptance criteria, architectural conformance, module scope, Reflect completeness, security (OWASP patterns), correctness (9 named bug patterns with explicit blocking thresholds), and code quality. Produces an ephemeral advisory report with tiered findings (Blocking / Advisory) and a MERGE-READY / NEEDS-WORK verdict. Supports feature branches (Full mode) and fix/tweak branches (Lightweight mode).
• SKILL.md: Added Code Review to Agent Operations, Agent Autonomy Boundaries table, Builder Commands ("Code review", "Review feature", "Review fix", "Review tweak"), and quick reference table.
• Lint fixes: Resolved pre-existing ruff errors in history.py (timezone.utc → UTC, two long CSS lines) and tests/test_init.py (unused import).

Version 0.5.0

• Lifecycle & PRs: Per-initiative lifecycle.json (in drafts/active) defines PR boundaries (specs, initiatives, features, tasks) and an ordered step list. Created during Approach via create_lifecycle.py; promoted at kickoff and archived with the initiative.
• Status merge detection: When lifecycle.json exists, status.py reports which branches are merged (git-based) and suggests the next lifecycle step (Merged/Next).
• open_pr.py: New script to open a PR from the current branch (tries gh → glab → Bitbucket URL → fallback). Host-agnostic; no API keys.
• SKILL/HOW-TO/CLAUDE: Updated Complete feature/initiative for PR-at-boundary wording, lifecycle step 5b, and create_lifecycle/open_pr in CLI references.
• Requirement change: Cicadas now requires Python 3.11+ (was 3.13+). The scripts use stdlib only (e.g. datetime.UTC); the installer and docs have been updated accordingly.

Version 0.4.0

• New Feature: Added install.sh — a single-command bash installer that checks for Python 3.12+, downloads Cicadas from the GitHub archive URL, extracts to src/cicadas/ (configurable via --dir), and calls init.py to bootstrap .cicadas/.
• Agent Integrations: --agent flag supports claude-code (symlink at .claude/skills/cicadas), antigravity (symlink at .agents/skills/cicadas), and cursor (copies SKILL.md to .cursor/rules/cicadas.mdc). Interactive agent prompt when stdin is a tty; skipped gracefully in curl | bash context.
• Update Workflow: --update flag re-downloads and overwrites skill files without touching .cicadas/ state or re-running init.
• Docs: README.md and HOW-TO.md updated with one-liner install, agent integration table, and update workflow.
• Tests: Added 2 tests to test_init.py covering hook installation loop and __main__ entry point; init.py coverage 74% → 97%. Overall test coverage 83% → 84% (54 tests).
• Canon: Updated product-overview.md and tech-overview.md with installer feature, distribution architecture, and agent integration conventions.

Version 0.3.5

• Bug Fix: Fixed archive.py leaving orphaned branch entries in registry.json after archiving an initiative; associated branches (matching initiative field) are now deregistered automatically.
• Tests: Added regression test test_archive_initiative_deregisters_associated_branches in test_archive_status.py.

Version 0.3.4

• New Script: Added src/cicadas/scripts/abort.py — context-aware escape hatch that detects the current branch type (tweak/, fix/, feat/, initiative/) and rolls back the appropriate scope, prompting whether to move promoted active specs back to drafts or delete them entirely.
• Skill Update: Added "Abort" builder command and abort.py entry to the CLI Quick Reference in SKILL.md.
• Docs: Updated README.md, HOW-TO.md, src/cicadas/README.md, and CLAUDE.md to document the new command.
• Tests: Added tests/test_abort.py with 14 tests at 86% coverage.
• Dev Dependencies: Added pytest and pytest-cov to pyproject.toml.

Version 0.3.3

• Bug Fix: Fixed branch.py creating orphaned nested active directories (e.g. active/tweak/X) for lightweight fix/tweak branches; active dirs now consistently use the initiative name (active/{name}/).
• Test Infrastructure: Added tests/conftest.py to ensure scripts dir is in sys.path before test collection, fixing test isolation failure.
• Tests: Added two regression tests for the active dir convention in test_branch.py.

Version 0.3.2

• New Script: Added src/cicadas/scripts/history.py — generates a self-contained HTML timeline (.cicadas/canon/history.html) from archived specs and the index ledger.
• Skill Update: Added "Project history" / "Generate history" builder command and history.py entry to the CLI quick reference in SKILL.md.

Version 0.3.1

• Claude Code Integration: Added CLAUDE.md with build/test/lint commands and codebase architecture overview.
• Skill Registration: Created .claude/skills/ directory with a symlink to src/cicadas/, registering Cicadas as a native Claude Code skill.
• SKILL.md Compliance: Renamed skill.md to SKILL.md per Anthropic convention; fixed missing opening --- frontmatter fence.
• SKILL.md Improvements: Added allowed-tools, argument-hint, and trigger-keyword-style description for reliable auto-invocation.
• SKILL.md Bug Fixes: Corrected duplicate emergence/ directory tree (scripts were misplaced), removed orphaned Bootstrap/Synthesis block, and added {cicadas-dir} definition note.

Version 0.3.0

• Lightweight Paths (Bugs & Tweaks): Introduced streamlined workflows for simple bug fixes and small tweaks: fix/ and tweak/ branches (fork from main), minimal specs (buglet.md, tweaklet.md), and emergence subagents (Bug-fix, Tweak). Scripts updated to support relaxed validation, direct-from-main branching, and status/history for fix/tweak.
• Test Coverage: Achieved >75% code coverage (83% overall) for all core scripts.
• Mock-Free Testing: Refactored the test suite to use real file system operations and Git scaffolding instead of mocks.
• Linting & Formatting: Integrated ruff and pre-commit hooks for automated code quality checks.
• Refactoring: Renamed signal.py to signalboard.py to resolve Python standard library collisions.
• Reliability: Fixed internal bugs in update_index.py and improved JSON serialization for Path objects.

Version 0.2.1

• Renamed the core skill directory from scripts/chorus/ to src/cicadas/.
• Created an Antigravity skill symlink in .agents/skills/cicadas pointing to the new src/cicadas/ location.
• Updated documentation (README.md) to reflect the new paths and branding changes from Chorus to Cicadas.
• Updated all skill files to use relative paths within the skill.
• Created a comprehensive README instruction file