Repository Issue Activity (beta)

#70 SessionStart always-on hook errors on Windows: 'sh' and ${CLAUDE_PLUGIN_ROOT} don't resolve under PowerShell

#96 Misinterpreted "Cap lists at 5 items"

#93 Bug: Nested SKILL.md metadata breaks strict Agent Skills parsers

#98 evals: `agent-owned-edit` cannot be passed by any run

#91 hook codex Support

#90 Name too long

#99 Rule 8 may pressure unfounded cause attribution (partial-success regression)

#4 Benchmarks on whether this reduces performance of the model

#64 i have adhd

#92 feat: add caveman-style grammar compression as optional layer

#42 Does answer-first structure trade off against language quality?

#75 GTD

#6 Can we have plugins for OpenCode?

#59 INSTALL.md: no GitHub Copilot section, though Copilot now reads Agent Skills natively

#43 Several rules fight agentic-harness system prompts; tangent suppression filters at the wrong layer

#61 manifest.hooks duplicates the auto-loaded hooks/hooks.json — SessionStart hook never loads on Claude Code

#62 Plugin fails to load in Claude Code: duplicate hooks/hooks.json registration

#3 Suggestion: mention to use the task tools (claude) or plan tool (codex) to divide walls of texts into chunks and track them as an internal todolist

#55 .cursor skill is a symlink — broken on Windows clones and GitHub ZIP downloads

#54 "How activation works" claims the model can never self-apply the skill — false for Codex

#52 Claude eval runner inherits the operator's global config — the always-on flag injects the skill under test into the baseline condition

#53 Unmetered-runner rejection is raised after the first paid call, though it is knowable before any call

#47 INSTALL.md "Want different rules" still omits the install step

#44 Release gate passes when baseline and candidate were judged on different cases

#45 `--case` with an unknown id runs nothing and exits 0