Skip to content

CHANGELOG.md

Latest commit

 

History

History
383 lines (265 loc) · 29.3 KB

CHANGELOG.md

File metadata and controls

383 lines (265 loc) · 29.3 KB
Raw

Changelog

Unreleased

  • Run the Bash installer suite on Ubuntu and the PowerShell installer suite on both Ubuntu and Windows through a dedicated installer-tests GitHub Actions workflow. The existing shared-rule drift workflow remains separate.

1.1.0 - 2026-07-05

First release after v1.0.0. Installer replacement, docs/ai/ lifecycle rework, and behavioral-eval hardening, plus incremental skill wording improvements carried over from post-v1 maintenance.

Installer replacement:

  • Add one canonical, static package inventory for reusable Claude, reusable Codex, and Claude bootstrap installation, with LF-normalized SHA-256 content validation before and after copy.
  • Add explicit normal install, overlay reinstall (--force / -Force), and staged clean reinstall (--clean / -Clean) behavior across Bash and PowerShell installers. Conflicting flags and managed path type conflicts fail; overlay preserves and reports extra files.
  • Add installer behavior tests covering both managed packages, including a Bash test suite (scripts/test-installers.sh) mirroring the existing PowerShell coverage.
  • Fix a Bash installer defect where a successful normal install or --clean reinstall printed a success message but exited with status 1, because the EXIT trap's cleanup check used [ -d "$stage" ] && rm -rf "$stage" as a bare statement; a false test in that position is exempt from set -e but still sets the trap function's own exit status, which becomes the script's final exit code. Every Bash wrapper script (install-user.sh, install-project.sh, install-codex-user.sh, install-codex-project.sh, install-bootstrap-user.sh) ends by calling install-package.sh as its last statement, so this previously caused those default-mode and --clean invocations to report failure to callers despite installing correctly. Fixed by using explicit if/fi instead of && for both the trap cleanup and the post-replace backup removal. Caught by the new Bash test suite.

docs/ai/ lifecycle rework:

  • Establish the fixed docs/ai/ package with ORIENTATION_STATE.json, manifest-backed freshness and provenance, full and scoped reconciliation, exact-current reuse, legacy adoption approval, dirty-tree handling, and no-Git behavior.
  • Retire automatic generation and mutation of project-local .claude/skills/codebase-orient/SKILL.md without adding an overlay or migration subsystem.

Behavioral-eval hardening:

  • Extend the existing behavioral-evaluation runner with the authored no-date-only-churn and source-drift two-pass cases and an optional model override.
  • Fix the behavioral-evaluation runner's Codex sandbox isolation: CODEX_HOME was left pointing at the real user's .codex directory even though HOME and USERPROFILE were overridden to a disposable sandbox, and the disposable .codex directory was never created. Both meant Codex could read real global state during a nominally isolated eval run. Fixed by pointing CODEX_HOME at the disposable home and creating that directory before invocation. A fully empty disposable CODEX_HOME cannot authenticate, so the runner now seeds it with a copy of the real auth.json only - no other real Codex state (skills, session history, config) is copied in.
  • Validation status for the two named two-pass cases: the isolation and authentication fix above was confirmed against a real codex exec run (the observed failure moved from a missing-directory error, to 401 Unauthorized, to a real account usage-limit response, then to a real completed run once the account quota reset). A full live pass/fail verdict for either case is still not available: in the completed run, the Codex session reported itself as read-only despite the case requesting workspace-write, so pass 1 correctly proposed the package instead of writing it, per the skill's own dry-run fallback rule, rather than exercising the intended write-then-rerun-unchanged check. This is recorded as a known Codex CLI environment limitation, not an Orient skill defect; unlike v1.0.0, this release does not carry an independent human-through-agent validation pass.

Orientation skill improvements (canonical + bootstrap):

  • Treat existing docs/ai output as hypotheses on re-run: the staleness rule now explicitly applies the docs-as-hypotheses discipline to prior orientation output, preventing stale structural claims from being accepted as discovery input on subsequent passes.
  • Add agent-closable vs human-required classification axis to open questions: the orientation-completion-rule now requires each unresolved question to be classified not only by urgency (Blocking / Relevant-but-non-blocking / Background) but also by who can act on it, letting a new agent immediately identify which items it can work on autonomously.
  • Surface check script constraints in CHANGE_SURFACES docs-impact guidance: the change-surfaces-mapping-guidance rule now instructs orient to note what file types or patterns scripts/ check scripts enforce, so future agents know about constraints like ASCII-only requirements before writing.
  • Add conditional release status section to CODEBASE_MAP: the discovery order now instructs orient to add a Release status section to CODEBASE_MAP.md when a release plan, milestone doc, or version changelog is found, summarizing current version, active milestone, and blockers.
  • Clarify that docs/ai files should be committed in target repos: the output files section now states explicitly that these files are authored narrative that improves across sessions and should not be gitignored in target repos.
  • Flag release validation records and gate scripts as a hidden-risk surface: the hidden-risk-reporting-rule now includes release validation records and gate scripts in its checklist, and instructs orient to note in CHANGE_SURFACES that file changes after a validated candidate require a re-check before tagging.

Post-v1 maintenance hardening (no installer behavior change at the time):

  • Add a deterministic shared-rule drift guard for canonical skills/codebase-orient/SKILL.md vs the bootstrap embedded shared-rule snapshot in skills/install-codebase-orient/SKILL.md, including explicit shared-block markers, a validation script, and a GitHub Actions check that now guards those shared blocks against future accidental drift.
  • Align the bootstrap embedded shared-rule snapshot to canonical wording so future bootstrap-generated project-local skills receive the synchronized shared-rule content.
  • Freeze the former docs/V1_RELEASE_PLAN.md contents as docs/releases/v1.0-validation-record.md and keep docs/V1_RELEASE_PLAN.md as a short compatibility pointer now that v1.0.0 has shipped.
  • Add an initial local/manual Codex behavioral eval scaffold: an inspectable prompt corpus plus a PowerShell wrapper over a Node runner that uses isolated disposable fixtures, stores traces outside the repo by default, and currently emits evidence summaries for one validated single-case vertical slice with proxy-only invocation evidence from local codex exec --json traces.
  • Clarify that the reusable canonical skill currently has an explicit tuned framework-probe section for SvelteKit, while other frameworks still rely on the generic discovery order unless later live-fire or eval evidence justifies dedicated tuned probes.

1.0.0 - 2026-05-24

First stable public release of codebase-orient.

  • Publish the reusable repository-orientation skill for Claude Code and Codex with guarded user-level and project-local install paths.
  • Keep Claude Code bootstrap support available as a Claude Code only path.
  • Preserve the agent-delegated installation contract: exact tagged source acquisition, checked-in installer execution, no reconstructed SKILL.md, and stop-and-ask fallback when exact installation is not possible.
  • Keep explicit invocation guidance and the documented authority boundaries for generated docs/ai/* orientation artifacts.
  • Include the public SECURITY.md reporting policy and the completed public-release/governance preparation record.
  • Record that independent human-through-agent validation passed against public v1.0.0-rc.6.
  • No marketplace or plugin packaging is part of this release.

1.0.0-rc.6 - 2026-05-23

Candidate for rerunning independent human-through-agent validation. No orientation skill behavior change. No installer target, copy, or overwrite semantics change.

Exact-install integrity for agent-delegated onboarding:

  • Add a compact README rule that agent-delegated installation must start from an exact local copy of the requested tagged source and must use the checked-in installer script for the selected tool and scope.
  • Clarify that project-local delegated installation is valid when requested, but it must still use an exact tagged checkout/archive plus the checked-in installer rather than reconstructed skill content.
  • Explicitly forbid manually recreating, pasting, summarizing, paraphrasing, or rewriting SKILL.md from web-fetched, rendered, transformed, or partial content; if exact acquisition or installer execution is not possible, the agent must stop and ask.

External validation evidence:

  • Record that an independent v1.0.0-rc.5 human-through-agent attempt failed the external validation gate on installation integrity.
  • In that run, a new Claude Code session reasonably chose the supported project-local route from a minimal public tag handoff, but then began an approximate manual-install path that would have reconstructed installed skill content from fetched non-authoritative material.
  • The attempt was interrupted before any installed skill file was written; only empty install directories were created in the target project, and no tracked target-project change was observed.

Release status:

  • v1.0.0-rc.5 remains the prior public agent-delegated onboarding candidate and historical evidence for the failed exact-install integrity gate.
  • v1.0.0-rc.6 is the active candidate for rerunning independent human-through-agent validation.
  • Final v1.0.0 remains blocked until that external validation gate passes against published rc.6 and the final release checklist is rerun.

1.0.0-rc.5 - 2026-05-23

Release candidate for independent human-through-agent validation. No orientation skill behavior change. No installer target, copy, or overwrite semantics change.

Onboarding and security:

  • Add a first-class agent-delegated installation path to README.md, including a pasteable prompt that tells a coding agent to follow the documented install contract, avoid silent overwrite, report explicit invocation, and stop before running orientation in a target repo.
  • Clarify existing-install decisions in README.md: overlay refresh (-Force / --force), clean reinstall (delete first), and intentional project-local install.
  • Clarify the distinction between installation, Codex skill discovery, explicit invocation, and a later orientation run. Reading an installed SKILL.md is described as file-copy evidence, not proof that the tool discovered or invoked the installed skill.
  • Add a root SECURITY.md that points reporters to GitHub private vulnerability reporting, with email fallback.

Codex installer wording alignment:

  • Update Codex installer output only where needed to direct users to explicitly invoke codebase-orient after installation using the documented Codex skill-selection flow.
  • Align README and Codex installer wording to current official Codex documentation: Codex detects newly installed or updated skills automatically; if codebase-orient does not appear, restart Codex; and in Codex CLI/IDE the explicit invocation path is /skills or $.

Release status:

  • v1.0.0-rc.4 remains the prior published sanitized candidate and is unchanged.
  • v1.0.0-rc.5 is the prior public agent-delegated onboarding candidate.
  • Final v1.0.0 remains blocked until that revised external validation gate passes and the final release checklist is rerun against the published candidate.

1.0.0-rc.4 - 2026-05-23

Pre-public sanitation candidate for external cold-user validation. No skill behavior or installer changes.

Documentation sanitation:

  • Remove maintainer-local filesystem paths from tracked public-intended docs.
  • Generalize private validation target names and local snapshot identifiers to public-safe repo-type descriptions while preserving the validation outcomes that matter for release confidence.
  • Keep the public distribution model unchanged: documented manual installation only; no marketplace or plugin packaging work added.

Release status:

  • v1.0.0-rc.3 remains the last pre-public contract-correction candidate, but it is blocked for publication because the tracked docs and reachable tags/history still carried private validation identifiers and local path disclosures.
  • v1.0.0-rc.4 is the new publication-safe candidate for independent cold-user validation.
  • External cold-user validation remains pending and must now run against v1.0.0-rc.4.

1.0.0-rc.3 - 2026-05-23

Pre-validation contract correction before external cold-user testing. No skill behavior or installer changes.

README corrections:

  • Narrow the early safety NOTE: distinguish normal-mode writes (docs/ai/* creation or refresh) from dry-run mode (proposes changes without writing); disclose that Claude Code project-local installs may also update .claude/skills/codebase-orient/SKILL.md with verified project-specific discovery paths.
  • Add ### Claude Code project-local specialization under ## Important notes: concisely states what specialization writes, that canonical rules are not overwritten, that dry-run mode suppresses the write, and that Codex project-local installs do not use this behavior.
  • Restore a compact ## Limitations and trust section: explicit invocation is the reliable path; orientation improves process not correctness; large monorepos may need scoped orientation; skills are agent instructions and SKILL.md should be reviewed before installing.

Release status:

  • v1.0.0-rc.1 and v1.0.0-rc.2 remain intact as immutable historical candidates.
  • v1.0.0-rc.2 should no longer be used for external cold-user validation; the mutation-scope and trust-posture wording corrected here was not present in that candidate.
  • At the time of v1.0.0-rc.3, external cold-user validation was intended to run against that candidate. That intent was later superseded by v1.0.0-rc.4 after the pre-public sanitation pass.

1.0.0-rc.2 - 2026-05-23

Promote the post-rc.1 user-facing corrections into the active release candidate for external cold-user validation.

User-facing corrections:

  • Rewrite README.md around a faster newcomer journey: what the skill is, which install path applies, the exact command to run, the first prompt to give the agent, and the expected repo changes.
  • Clarify local acquisition and shell-script invocation: README now supports clone or download flows and uses bash ./scripts/... or bash /path/to/... for macOS/Linux examples where archive downloads may not preserve executable bits.
  • Fix project-local .gitignore guidance emitted by the Claude Code and Codex project-local installer scripts: all four scripts now print .claude/* or .agents/* with the required negations so the shared SKILL.md file can actually be re-included and tracked.
  • Behaviorally verify the corrected .gitignore guidance with git add -n / ignore-state checks before promoting this candidate.

Release status:

  • v1.0.0-rc.1 remains intact as the prior historical candidate and does not contain these README / installer-guidance fixes.
  • Final v1.0.0 remains blocked on the external cold-user validation gate and should be tested against v1.0.0-rc.2, not v1.0.0-rc.1.

1.0.0-rc.1 - 2026-05-22

Freeze the cold-user validation baseline. All implementation and accumulated validation evidence are complete. Final v1.0.0 tag is gated by one remaining external acceptance result: an independent cold-user install simulation (G.2).

Validation and evidence (docs only):

  • Add docs/V1_RELEASE_PLAN.md: v1.0 definition, release criteria, test matrix, live-fire validation matrix, blockers, non-goals, known risks, definition of done, and next immediate step.
  • Record completed v0.3.2 install matrix validation: Windows PowerShell (5 scripts, 5 cases each) and Git Bash (18 checks, 18/18 PASS) from disposable tagged clone.
  • Complete canonical/bootstrap embedded-template drift check: no material unintentional drift found; no edits to either SKILL.md required.
  • Backfill live-fire validation evidence from prior Claude Code passes: one external SvelteKit/frontend repo (normal mode) and one external Laravel/backend/deployment-sensitive repo (normal mode). F-matrix Rows 1, 2, 7, 8 satisfied.
  • Record external Codex live-fire pass in the external Laravel/backend/deployment-sensitive repo (project-local skill, cross-agent cache lifecycle, no date-only churn, no application code modified). Close G.4. F-matrix Row 6 PASS.
  • Record a second cross-agent cache-lifecycle validation in an external SvelteKit/static-portfolio/deployment-workflow repo: Claude Code refreshed orientation cache; Codex found substantive README drift and pushed only documentation/orientation corrections. Corroborating second external cross-agent lifecycle evidence.
  • Record blind no-Git dry-run validation in an unseen legacy PHP CMS snapshot (Codex user-level skill, no prior docs/ai). Codex surfaced a real maintenance-doc versus config contradiction and recommended dry-run only. F-matrix Row 4 PASS. F-matrix rows satisfied: 6/8.
  • Polish README onboarding clarity before independent cold-user validation: surface the default install choice and make update guidance explicit without changing behavior.

No Skill or installer behavior changes.

0.3.2 - 2026-05-22

Normalize and enforce ASCII punctuation in tracked repo-maintenance text.

Smart punctuation removal:

  • Replace ~210 lines of em dashes across CHANGELOG.md, README.md, and both SKILL.md files.
  • Replacement rules: closing markdown span (**label**, _label_, `code`) followed by em dash converted to colon-space; all other em dashes converted to space-hyphen-space.
  • En dashes in numeric ranges (3-5) replaced with hyphen-minus.
  • <=150 replaced with under 150 in CHANGELOG.

Enforcement:

  • Add scripts/check-ascii-punctuation.ps1 and scripts/check-ascii-punctuation.sh - scan tracked text files for forbidden smart punctuation; exit 1 if found.
  • Update README development notes: strict ASCII-only convention now covers all tracked repo-maintenance text, not just terminal output strings. SKILL.md files are read as style templates by Claude/Codex; smart punctuation bleeds into generated orientation docs.

No skill or installer behavior changes.

0.3.1 - 2026-05-22

v1 contract cleanup based on skeptical second-opinion audit.

Artifact policy - .gitignore:

  • Add .agents/ to .gitignore: Codex project-local install artifacts in this source repo are local runtime artifacts and should not be tracked.
  • Add docs/ai/ to .gitignore: generated self-orientation cache is non-canonical in this source repo and should not be tracked.

Install refresh semantics - README.md:

  • Document that -Force / --force is an overlay install: source files are copied over the target, but files removed from the source package since the last install are not pruned from the installed target.
  • Add delete-first workaround note for clean exact-sync reinstalls.

Project-local CWD contract - README.md:

  • Add explicit note that project-local install scripts use the current working directory as the install root and do not verify it is a repository root.

Codex lifecycle parity - README.md:

  • Add callout after Codex project-local install section: bootstrap skill and post-orientation project-local specialization are Claude Code only features; Codex project-local installs get the generic skill and require explicit invocation.

Manual copy example alignment - README.md:

  • Change bash manual copy examples from cp -R .../\* to cp -r .../. to match script semantics (includes hidden files; currently harmless since the package has none, but examples now match exactly).

0.3.0 - 2026-05-21

Consolidates post-v0.2.0 work into a tagged release. The orientation behavior additions recorded in the v0.2.1 development entry (instruction-layer topology, CHANGE_SURFACES mapping guidance, agent handoff summary) are included under this tag - v0.2.1 was a named development checkpoint that was never separately tagged.

Installer output hardening - scripts/*.sh, README.md (9bc01df):

  • Replace non-ASCII characters (em dashes, curly quotes) in installer script terminal output with ASCII equivalents.
  • Reduces mojibake rendering in PowerShell 5.1 / Windows terminals on code page 1252.
  • Added ASCII-only output encoding convention to README development notes.
  • No install behavior changes.

Authority-boundary clarification - skills/codebase-orient/SKILL.md (a707995):

  • Add thin-overlay framing to project-local specialization rule: project-specific paths are the thin customisable layer; canonical rules are the stable layer across all repos; keep them visually separated.
  • Closes source/mirror inconsistency introduced in c75c675.

Authority-boundary clarification - skills/install-codebase-orient/SKILL.md (c75c675):

  • Fix Last refreshed: contradiction in idempotency rule: now defers to the no-date-only-churn rule.
  • Add generated-cache framing note after creates table: docs/ai/ files are orientation aids; source code and project config remain authoritative.
  • Add thin-overlay framing to project-local specialization rule in embedded template.
  • Add SvelteKit subdirectory note and critical server-side route files callout to bootstrap discovery probes, matching canonical skill.
  • Add internal changelog entries for v0.2.0 and v0.2.1 syncs (previously unrecorded).
  • Bump install-codebase-orient frontmatter version to 0.2.2.

0.2.1 - 2026-05-21

Weaves orientation-core deep-research findings into the skill without adding new default docs files, hooks, or automation.

Findings woven in (canonical skill + bootstrap embedded template):

  • Instruction-layer topology: discovery step 1 now notes which instruction files are present (CLAUDE.md, AGENTS.md, .claude/settings.json) and at which scope (project-local vs user-level); recorded in CODEBASE_MAP.md for agent handoff clarity.
  • CHANGE_SURFACES mapping guidance: new section explicitly calls out three additional surface categories: auth/admin/operator UX changes (with a11y/WCAG note), deployment-sensitive changes (with smoke-check entry-point note), and docs-impact changes (which docs need updating per subsystem). Hard rule added: no separate smoke-check or handoff files in docs/ai/.
  • Agent handoff summary: orientation report discipline now includes an optional compact handoff block (snapshot, blocking questions, critical surfaces, recommended first action; under 150 words, no new file).

Bootstrap sync: skills/install-codebase-orient/SKILL.md updated - CHANGE_SURFACES template section and embedded downstream template both reflect the three new findings.

Rejected: hooks, CI automation, release checklist, full a11y audit, separate AGENT_HANDOFFS.md / DEPLOYMENT_SMOKE_CHECKS.md by default, safe-cleanup behavior, sceptical review behavior - none are orientation-core.

0.2.0 - 2026-05-21

Dual-runtime orientation skill with bootstrap support. Adds Codex install, recursive skill copy, user-level bootstrap install scripts, live-fire-driven refresh behavior, cross-file consistency rules, no-date-only-churn rule, and trigger-oriented / token-aware invocation guidance.

No-date-only-churn fix from second no-wrapper live-fire test:

  • No-date-only-churn rule: do not rewrite a docs/ai/ file solely to bump the Last refreshed date. A verified-current doc is left unchanged. Last refreshed is updated only when content changes for a substantive reason.
  • Orientation report discipline: final report must label each docs/ai/ file as: Created / Substantively updated / Verified current (unchanged) / Proposed only / Skipped. Reporting a date-only rewrite as "updated" is not allowed.
  • Bootstrap embedded template synced: both rules added to skills/install-codebase-orient/SKILL.md.
  • No installer behavior changes.

Invocation reliability and token-aware use improvements:

  • Trigger-oriented skill description: codebase-orient frontmatter description rewritten to be richer in trigger signals: new/unfamiliar repo, new session, before multi-file work, before refactor/planning/handoff, stale docs/ai/, unclear structure. Includes explicit trigger phrases (scan, orient, survey, familiarize, before I start, etc.) and explicit skip signal (tiny single-file known edits).
  • When-to-use / when-to-skip guidance: "When to use this skill" section split into explicit use cases and explicit skip cases. Replaces the previous flat bullet list of task types.
  • Token-aware orientation cache guidance: new "Token-aware use guidance" section explaining docs/ai/ as an orientation cache: use when cost amortizes over broad work or sessions; skip for tiny known tasks; do not save tokens by guessing instead.
  • Optional AGENTS.md / CLAUDE.md snippet: README now includes recommended project-instruction snippet for both Codex (AGENTS.md) and Claude Code (CLAUDE.md). Marked optional; does not guarantee invocation; triggers when useful, suppresses when not needed.
  • Bootstrap embedded template synced: when-to-use/skip and token-aware guidance added to skills/install-codebase-orient/SKILL.md embedded downstream template. Also added "audit only" / "no writes" to dry-run triggers.
  • No installer behavior changes.

Live-fire findings from no-wrapper /codebase-orient test applied.

  • Normal/dry-run mode surfaced earlier: moved to immediately after Hard rules, before discovery and output rules, so it is visible without scrolling through the full skill. Added "audit only" and "no writes" as recognized dry-run triggers.
  • Cross-file consistency rule: new rule requiring CODEBASE_MAP.md, CHANGE_SURFACES.md, and OPEN_QUESTIONS.md to remain coherent with each other. Covers resolved questions, new surfaces, new uncertainty, and direct contradictions. Applied to both canonical skill and bootstrap embedded template.
  • SvelteKit route probing expanded: server-side route files (+page.server.ts, +layout.server.ts, +server.ts) now listed explicitly alongside +page.svelte and +layout.svelte in the canonical skill and bootstrap discovery section. Note added that server-side files are critical for auth, data loading, form actions, and API endpoints.
  • Formatter fallback guidance: formatter guidance now explicitly handles the case where a formatter is missing, unavailable, or not configured for Markdown. Skip with a note; do not treat missing formatter support as an orientation failure.
  • Read-depth clarification: new "Path existence vs content read" subsection clarifies when path existence alone is sufficient (low-risk inventory claims) versus when actual file content must be read (claims affecting behavior, architecture, risk, auth, routing, or change surfaces). Applied to both canonical skill and bootstrap embedded template.
  • Project-local specialization path made explicit: target path .claude/skills/codebase-orient/SKILL.md now stated explicitly. Added note that this applies only when a repo-local Claude Code skill exists or is being generated, and that Codex installs are handled separately. Applied to both canonical skill and bootstrap embedded template.
  • Bootstrap embedded template synced: all six findings reflected in skills/install-codebase-orient/SKILL.md.

0.1.3

  • Added scripts/install-bootstrap-user.ps1 and scripts/install-bootstrap-user.sh: user-level Claude Code install scripts for the install-codebase-orient bootstrap skill.
  • Updated README with bootstrap skill install instructions (user-level, Claude Code only).
  • No project-local bootstrap install scripts added; no Codex bootstrap install support added.

0.1.2

  • Added skills/install-codebase-orient/SKILL.md: tracked source for the Claude Code bootstrap skill. Bootstrap skill source is now versioned here but is not yet part of the automated install flow.
  • Updated README with "Skills in this repo" section listing both skill sources and their tool targets.

0.1.1

  • Added Codex install support: scripts/install-codex-user.ps1, install-codex-project.ps1, install-codex-user.sh, install-codex-project.sh.
  • Single shared skill source skills/codebase-orient/SKILL.md unchanged - install scripts copy it to the appropriate target path for each tool.
  • All install scripts now copy skill contents recursively, preserving optional subdirectories (future scripts/, references/, assets/ etc).
  • README now positions the project as an Agent Skill for both Claude Code and Codex.
  • README now documents install paths: Claude Code uses .claude/skills/, Codex uses .agents/skills/.
  • README now explains invocation reliability: auto-invocation is model-driven and not guaranteed; explicit invocation is preferred.
  • README now documents AGENTS.md as the Codex persistent instruction layer equivalent to CLAUDE.md.
  • README now includes .gitignore guidance for Codex project-local installs.
  • README now includes Codex uninstall instructions.

0.1.0

  • Initial private release.
  • Added codebase-orient Claude Code skill.