fix(ci): repair pre-existing main failures (source-tests + MegaLinter)#110
Merged
Conversation
…tension
Replace the local-render approach (pi-request-dump.mjs) with a real wire
capture using the same Anthropic /v1/messages loopback as Claude Code.
How it works:
- audit_pi() writes a throwaway PI_CODING_AGENT_DIR containing:
- lib/pi-loopback-ext.mjs an ES module extension that overrides the
built-in 'anthropic' provider baseUrl to http://127.0.0.1:$PORT
- settings.json loads the extension; dummy Anthropic model
- auth.json (0600) dummy API key so pi doesn't bail pre-request
- Starts claude-loopback.mjs (Anthropic messages format)
- Runs: PI_CODING_AGENT_DIR=... PI_AUDIT_LOOPBACK_PORT=... PI_OFFLINE=1
ANTHROPIC_API_KEY=audit-dummy pi -p 'hi' --approve
- Captures the full /v1/messages body including the live extension stack
output (<available_skills> block is now present when skills are installed)
- Uses claude_first_turn_chars for FIRST_* columns (Anthropic format)
SYS_* column is unchanged (pi-dump.mjs local render, no network).
Before: pi FIRST_CH ~4,969 (local render, no skills, no extension stack)
After: pi FIRST_CH ~24,096 (wire capture, skills catalogue included)
Also update categorizePi in both categorize.mjs and breakdown-unified.mjs to
read the Anthropic wire format: system array -> systemText, span breakdown
now includes <available_skills> anchor, tool chars from req.tools array.
Uncommented yubico-authenticator from disabled packages list. It will now be installed on macOS via brew bundle. Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
Bind Ctrl+1..Ctrl+5 to jump directly to the Nth Space on the focused screen, including full-screen-app spaces that the native "Switch to Desktop N" shortcuts cannot reach. Uses hs.spaces.spacesForScreen plus hs.spaces.gotoSpace, which drives the Mission Control thumbnail button. Installs Hammerspoon via the Homebrew cask (macOS only) and gates the ~/.hammerspoon config to darwin in .chezmoiignore. Config auto-reloads via hs.pathwatcher. One-time manual step: grant Hammerspoon Accessibility permission in System Settings -> Privacy & Security -> Accessibility. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
macOS reserves Ctrl+1..Ctrl+N as Carbon hotkeys for "Switch to Desktop" when N Mission Control user desktops exist, even when the System Settings shortcut checkboxes are off. RegisterEventHotKey then rejects those combos (error -9878), so hs.hotkey.bind silently failed for Ctrl+1/Ctrl+2 (two user desktops here) while Ctrl+3..5 registered fine. Replace the per-key hs.hotkey binds with a single hs.eventtap that reads Ctrl+number from the event stream before hotkey dispatch, so all five keys work uniformly and still reach full-screen spaces. Also enable hs.ipc in init.lua so the running config can be queried from the `hs` CLI (used to diagnose this). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add a Hammerspoon section to the repo guide: use hs.eventtap (not hs.hotkey.bind) for Ctrl+number Space switches because macOS reserves Ctrl+1..Ctrl+N for Switch-to-Desktop and RegisterEventHotKey rejects them (-9878); hs.spaces reaches full-screen spaces; hs.pathwatcher auto-reload survives chezmoi writes; hs.ipc CLI for debugging; Accessibility permission is a one-time manual grant. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Show the active Space's position (the same 1-based index Ctrl+1..Ctrl+5 use) as a menu bar badge via hs.menubar, updated on hs.spaces.watcher events. The dropdown lists every Space (full-screen ones flagged, current one checked) and clicking jumps to it. The badge is recomputed from spacesForScreen + activeSpaceOnScreen (not the watcher's deprecated space-number arg), deferred ~0.25s so it reads the settled position after macOS finishes the transition and any most-recent-use reordering. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Document the menu bar Space indicator (space_indicator.lua) and the macOS mru-spaces default that reorders full-screen spaces, which makes position- based Ctrl+N and the badge drift; record the defaults write to disable it and the watcher settle-delay workaround. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add home/.chezmoiscripts/run_onchange_after_macos_defaults.sh.tmpl to set `com.apple.dock mru-spaces -bool false` so macOS stops reordering Spaces by most-recent-use. That keeps the Hammerspoon Ctrl+1..Ctrl+5 direct-Space switches and the menu bar indicator pointing at consistent positions, including full-screen-app spaces. Darwin-guarded (renders empty elsewhere); run_onchange so it only re-runs when the content changes; restarts the Dock only when it is running so it is a no-op in headless/CI sessions. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Replace the small circled-number badge with one dot per Space on the focused screen, the active one filled (●) and the rest hollow (○), in Mission Control order. More legible at a glance and shows position within the full set of Spaces. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…empty When Ctrl+N targets a position that does not exist yet (for example Ctrl+3 with only two Spaces), add a new empty desktop with hs.spaces.addSpaceToScreen and switch to it instead of showing a "No Space N" alert. addSpaceToScreen does not return the new id, so the new Space is found by diffing the Space list before and after. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Match Komorebi's alt+number Workspace switches on Windows. On macOS Option+number normally types ¡™£¢∞, but the event tap returns true to swallow the keystroke before the input system inserts the character, so the Space switches and nothing is typed (verified by reading a focused TextEdit document after a synthetic ⌥1/⌥2: it stayed empty). Update the spaces.lua and space_indicator.lua comments and the AGENTS.md Hammerspoon section to reflect ⌥ and why the event tap is still required (special-char swallowing, plus Ctrl+number being unbindable via hs.hotkey). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
source-tests has been red on main since the bootstrap onchange hash landed: tests 8 and 9 (brewfile.bats) fail on ubuntu with `map has no entry for key "private"` at run_onchange_after_bootstrap.sh.tmpl:47. Cause: the render helper uses `chezmoi execute-template --init`, which does not promote the config template's [data] to top-level template vars, so each key the template reads must come via --override-data. The helper passed `codespaces` but not `private`; the bootstrap script gates a block on `.private` (derived from ~/.private in .chezmoi.toml.tmpl), so a fresh checkout (CI) errors while a dev machine passes because the key exists in on-disk config data. The production template is correct (real apply always sets .private); only the isolated test render under-specified its data. Reproduced and verified in an ubuntu container with chezmoi 2.70.4 (both tests pass after the fix). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…check MegaLinter (luacheck + stylua) has been red on main for the hammerspoon config: 'accessing undefined variable hs' and 2-space-vs-stylua-default formatting. hs is the Hammerspoon runtime global, and the watcher/tap/menubar are intentionally non-local globals (kept to avoid GC, per their comments). - Allow the hs global repo-wide via LUA_LUACHECK_ARGUMENTS in .mega-linter.yml. - Declare the intentional per-file globals (configWatcher, spaceSwitchTap, spaceIndicator*) with inline 'luacheck: globals' directives + a why comment. - Format all three files with stylua defaults (the repo has no stylua.toml). Verified locally: luacheck 0/0 and stylua --check clean on all three. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
audit.sh trips SC2016 (literal $schema JSON key inside single quotes) and SC2012 (ls on controlled Homebrew Cellar version dirs). Both are intentional; add scoped 'shellcheck disable' directives with reasons rather than rewriting. Verified: shellcheck --external-sources clean. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
devskim flags the JSON-Schema draft-07 $schema identifier (http://json-schema.org/draft-07/schema#) as an insecure URL in the consensus skill's *.schema.json. That URI is the canonical spec identifier and must stay http://, so it is a genuine false positive; add DS137138 to .devskim.json IgnoreRuleIds (alongside the existing DS162092). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
ChipWolf
changed the title
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Two checks have been red on
mainindependently of any feature work. This PR repairs both somain(and PRs branched from it) go green. None of these failures relate to a feature change; they're linter/test-harness issues on recently-merged code.1.
source-tests(ubuntu) —brewfile.bats8 & 9run_onchange_after_bootstrap.sh.tmpl:47reads.private, but the test renders viachezmoi execute-template --init, which doesn't promote the config template's[data]to top-level vars. The helper suppliedcodespacesbut notprivate, so a fresh checkout (CI) died withmap has no entry for key "private"while dev machines passed (key present in on-disk config data). The production template is correct (real apply always sets.private); only the isolated test render under-specified its data.Fix: add
"private":falseto the bootstrap render's--override-data. Verified in anubuntu:24.04container w/ chezmoi 2.70.4: both bats testsok.2.
MegaLinter— luacheck, stylua, shellcheck, devskimAll on recently-merged code (hammerspoon, agent-audit, consensus), not feature work:
home/dot_hammerspoon/*.lua): "accessing undefined variablehs" —hsis the Hammerspoon runtime global; the watcher/tap/menubar are intentional non-local globals (kept to avoid GC). Fix: allowhsviaLUA_LUACHECK_ARGUMENTS, declare the per-file globals with inlineluacheck: globalsdirectives.stylua.toml). Fix: format with stylua defaults.skills/agent-audit/audit.sh): SC2016 (literal$schemaJSON key) and SC2012 (lson controlled Cellar dirs) — both intentional. Fix: scopedshellcheck disabledirectives with reasons.skills/consensus/resources/*.schema.json): DS137138 flagshttp://json-schema.org/draft-07/schema#as insecure, but that's the canonical draft-07$schemaidentifier (must stayhttp://) — a genuine false positive. Fix: add DS137138 to.devskim.jsonIgnoreRuleIds.Verified locally:
luacheck0/0,stylua --checkclean,shellcheck --external-sourcesclean across all affected files. devskim is a config-only ignore (its SARIF showed only the four DS137138 false positives, no real findings); CI confirms.Relationship to #109
PR #109 (AGENTS.md → skills refactor) is clean on its own merits; its red CI is inherited from these two pre-existing main failures. Once this merges and #109 picks up
main, #109 goes green.🤖 Generated with Claude Code