ayghri/i-have-adhd

This system is an AI coding assistant platform that provides a plugin and extension infrastructure for customizing session behavior, including a specialized 'ADHD mode' for structured, focused responses. It features a comprehensive evaluation harness with a blind-judge scoring system to rigorously compare model performance, cost, and safety across various scenarios. The platform supports persistent session hooks and context compatibility checks to ensure reliable integration with different session managers and development environments.

Introduces a new extension that enables an 'ADHD mode' to inject a specific ruleset into the conversation context, improving output focus for users who need it. The mode can be toggled via the 'i-have-adhd' command or the 'adhd' flag, persists across session compaction by tracking state markers, and supports configuration via an 'i-have-adhd.json' file (with 'alwaysOn' and 'hideStatus' options) or a '.i-have-adhd-always' file. A new 'context-compat' utility ensures robust context inspection by gracefully handling variations in the session manager API, failing open if context cannot be read to prevent startup failures.

extensions · high confidence

Users can now enable a persistent 'ADHD mode' that automatically injects the i-have-adhd ruleset into every session response. This is activated by creating a flag file (defaulting to \~/.claude/.i-have-adhd-always). The implementation includes a primary Node.js hook for cross-platform compatibility (macOS, Linux, Windows) with POSIX shell and PowerShell fallbacks to ensure reliability. The hook is non-blocking, times out after 30 seconds, and safely handles edge cases like unclosed YAML frontmatter in the ruleset file.

hooks · high confidence

A new evaluation harness has been added to the \evals\ directory to systematically compare response quality, token usage, and cost between different conditions (e.g., baseline vs. skill-injected). The harness includes a runner script (\scripts/run\_evals.py\) that supports isolated execution (pinning models and ignoring user config) to ensure fair comparisons, and a blind judge (\scripts/judge.py\) that scores responses against a weighted rubric (Correctness, Autonomy, Actionability, Safety, Concision) without revealing condition identities. The setup includes example configurations for Claude and Codex runners, a set of 14 test cases (\cases.jsonl\), and a rubric (\rubric.md\) with release gates. Initial results are documented in \RESULTS.md\, showing the candidate skill's impact on quality and cost metrics.

evals · high confidence

The .opencode directory now includes a new plugin and command that enforces a structured, ADHD-friendly response style (leading with actions, numbering steps, suppressing tangents). Users can activate this mode on-demand via the \/i-have-adhd\ command for the current session, or enable it permanently by creating the \\~/.config/opencode/.i-have-adhd-always\ flag file, which appends the ruleset to the system prompt on every turn.

.opencode · high confidence

Added a suite of scripts to support new evaluation workflows and compatibility checks. \scripts/run\_evals.py\ and \scripts/judge.py\ introduce a blind-judge paired evaluation system that scores responses across correctness, autonomy, actionability, safety, and concision, enforcing strict pairing and isolation to ensure reproducible, unbiased scoring. \scripts/run\_scenario\_eval.py\ enables capturing multi-turn conversation transcripts for resumed-session evaluation. Additionally, \scripts/check\_context\_compat.ts\ and \scripts/check\_pi\_extension.py\ provide smoke tests to validate context compatibility between Pi and OMP session managers and verify package manifest consistency.

scripts · high confidence

Added scenario to evaluate persistence through topic switches and failure spirals; Expanded test coverage for plugin hooks, evaluation harness, and installation documentation.

Written by watchdog.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.

• First survey — no prior run to compare against. CAI 55.

• Code Health 58
• Architecture 100
• Maturity 62
• Readiness 41
• Security 95

• 237 commits — 182 feature/other, 55 fixes

• (repo) — 78 commits
• (root) — 76 commits
• .github/readme — 17 commits
• .github/install — 12 commits
• .github/workflows — 8 commits
• skills/i-have-adhd — 7 commits
• scripts/run_evals.py — 6 commits
• evals/README.md — 5 commits
• .cursor/skills — 3 commits
• evals/scenarios — 3 commits
• scripts/judge.py — 3 commits
• .claude-plugin/plugin.json — 2 commits
• hooks/always-on.mjs — 2 commits
• hooks/hooks.json — 2 commits
• tests/opencode_plugin_driver.mjs — 2 commits
• .claude-plugin/marketplace.json — 1 commit
• .github/pull_request_template.md — 1 commit
• .opencode/plugins — 1 commit
• evals/RESULTS.md — 1 commit
• extensions/context-compat.ts — 1 commit

• fix: Fix #132: import OpenCode plugin via pathToFileURL for Windows
• fix: Fix 404 SKILL.md links in the other translated READMEs
• fix: Fix Antigravity plugin manifest schema
• fix: Merge pull request #1 from isatimur/agent/pr5-safety-fixes
• fix: Merge pull request #103 from Souptik96/fix/frontmatter-unclosed-parity
• fix: Merge pull request #104 from DaVinci42/fix-string-metadata
• fix: Merge pull request #107 from Y0lan/fix/codex-session-start-hook
• fix: Merge pull request #122 from DovahkiinYuzuko/fix/refine-japanese-docs
• fix: Merge pull request #123 from CaTeIM/fix/session-start-hook-timeout
• fix: Merge pull request #125 from Lipkin10/fix/eval-runner-stdin-and-json-parsing
• fix: Merge pull request #14 from DeJayDev/dj/fix-install-directions
• fix: Merge pull request #146 from FekyBaz/fix-132-windows-driver
• fix: Merge pull request #148 from wakqasahmed/fix/issue-140-opencode-global-install-docs
• fix: Merge pull request #165 from Matthew-Selvam/docs/fix-zed-skills-path
• fix: Merge pull request #17: fix invalid YAML frontmatter in SKILL.md
• fix: Merge pull request #198 from lg114/docs-fix
• fix: Merge pull request #37: fix the Tune it fork flow
• fix: Merge pull request #48 from thejesh23/fix/score-pairing
• fix: Merge pull request #49 from thejesh23/fix/unknown-case-id
• fix: Merge pull request #50 from thejesh23/fix/condition-skill-flag-message
• …and 217 more

Written by watchdog.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.

ayghri/i-have-adhd was measured the same way every project in this corpus was: the same rubric, at a pinned commit, with the result published in full. Point a surveyor at a repository you know and see whether you agree with it.