test(github_intent): cover classification normalization edge cases#3
Draft
Koan-Bot wants to merge 1487 commits into
Draft
test(github_intent): cover classification normalization edge cases#3Koan-Bot wants to merge 1487 commits into
Koan-Bot wants to merge 1487 commits into
Conversation
Add defense-in-depth is_unlimited_quota() guards inside both _maybe_warn_burn_rate() and _downgrade_if_burning_fast() so burn-rate alerts and mode downgrades are suppressed even if the caller's budget_mode gate fails to resolve correctly.
…ent-1986 feat(skills): add /debug skill with 4-step hypothesis loop
…ss-burn-rate-unlimited-quota fix(burn-rate): suppress warnings when unlimited_quota is set
Add resolve_project_alias() fallback to 16 skill handlers that accept project names as arguments but only matched exact project names. Skills fixed: ai, deep, magic, branches, autoreview, brainstorm, incident, plan, deepplan, changelog, claudemd, done, explore, gha_audit, stats, diagnose. Pattern: exact name match first, then alias fallback — consistent with resolve_project_name_and_path() already used by audit, doc, profile, dead_code, spec_audit, private_security_audit, tech_debt.
…-project-alias
Move alias resolution before the show_perf early-return branch so /stats alias --perf correctly maps to the canonical project name. Previously the perf branch returned before alias resolution happened, causing empty results when using aliases.
Make is_unlimited_quota() resilient to malformed config (non-dict usage section) and fix the budget_mode exception fallback in plan_iteration() to respect unlimited_quota — previously defaulted to "session_only" even when unlimited_quota was set, causing spurious warnings/downgrades.
…ats-perf-alias fix(stats): resolve project aliases in --perf mode
…er-unlimited-quota-helper fix(quota): harden is_unlimited_quota() helper and budget_mode fallback
New status-group skill that displays current server date and time in a human-friendly format (e.g. 🕐 Tuesday June 17, 2026 — 16:15:42).
…me-skill feat(skill): add /time core skill with /date alias
Add post_threaded_reply() that threads replies to the original comment: PR review comments use GitHub's native in_reply_to threading, issue comments include a blockquote for visual context. Post brief acknowledgment replies when commands are queued from GitHub @mentions (e.g. "🤖 /review queued"), so users aren't left waiting with only a reaction. Configurable via github.ack_enabled (default on). Update /ask skill and natural-language reply paths to use threaded replies instead of top-level comments.
…ubmission Add an `approved` boolean (default: true) to the `review_verdict` config section. When disabled, review comments and PR feedback are still posted but the formal APPROVE/REQUEST_CHANGES status is not submitted via the GitHub API. The setting can be overridden per-project in projects.yaml via the `review_verdict` key under each project entry, following the same defaults-merge pattern as security_review.
Add `parallel_sub_commands` field to `Skill` dataclass, parsed from `parallel: true` in SKILL.md frontmatter. Introduce `ComboSkill` namedtuple as the return type of `collect_combo_skills()`, carrying both the command list and the parallel flag. Update `expand_combo_skill()` to batch-insert all sub-missions atomically via `modify_missions_file()` when parallel is set, while preserving the existing sequential insertion for non-parallel combo skills. Closes Anantys-oss#1988 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
New core skill that queues /security_audit, /dead_code, and /profile as parallel missions via the new parallel combo infrastructure. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add parallel combo skill documentation to skills README (frontmatter table + dedicated section with usage example and branch-independence warning). Add audit_all to user manual and skills reference. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Cover parallel batch insertion, URL-based dedup, no-project-tag handling, existing-mission preservation, ComboSkill namedtuple creation, and parallel flag parsing from SKILL.md. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add branch name to /status header: 'main - v0.76@abc +458' format. New get_branch() in version.py, graceful fallback when git unavailable.
…-show-branch feat(status): show current branch in header
…ent-1660 Implement: Expose usage, metrics, and log-tail endpoints in the REST A
…s warning Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…llama-launch) Removes the 'local' provider, its homegrown agentic runner, and every registry/onboarding/diagnostics/status reference. A stale cli_provider: local config now logs a clear startup warning and falls back to claude. ollama-launch is unchanged; the standalone 'make ollama' path keeps an 'ollama' sentinel so ollama serve still starts. BREAKING CHANGE: cli_provider: local is no longer supported; importing LocalLLMProvider from app.cli_provider/app.provider now raises ImportError. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…ent-1819 Remove deprecated local CLI provider (preserve ollama-launch)
Introduces a messaging.level setting with two values — debug (legacy firehose) and normal (new default: quiet, operator-focused). In normal mode the bridge stops narrating mission lifecycle, collapses GitHub/Jira per-mention chatter into a single aggregate count, and reports finished tracked skill missions as one scannable line with the PR URL. Failures and command replies still pass; every suppressed message is still logged. - messaging_level.py: env > state file > config.yaml > "normal" resolution + debug_only() gating helper (always logs, conditionally sends) - config.py: get_configured_messaging_level[_explicit]() - /messaging_level skill (alias /msglevel) to toggle live - run.py: concise normal-mode completion line; debug keeps verbose summary - loop_manager/jira: gate per-mention sends, emit aggregate counts - mission_executor: gate mission-start lines - startup_manager: one-time normal-default advisory (sentinel-gated) - docs + instance.example config Closes Anantys-oss#2064 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
feat(docker): recommend prebuilt GHCR image
…ent-2043 perf(quota): share one BurnRateSnapshot in /quota burn-rate line
…ent-2064 Implement: Reduce verbosity from Telegram / Slack
When a user configured Slack (KOAN_SLACK_* env vars) but left messaging.provider unset, resolution defaulted to telegram. The bridge then emitted 'Set KOAN_TELEGRAM_TOKEN and KOAN_TELEGRAM_CHAT_ID env vars.' and exited — despite a bridge platform being set up. resolve_provider_name() now infers a non-telegram provider when exactly one has credentials present (slack/matrix/discord via env, matrix/discord via config block). Explicit env/config still wins; ambiguous setups keep the telegram default. The spurious warning only fires when no platform is connected.
…og + tighten detection
…ing-provider-autodetect
When the Telegram bridge ran a background task (worker skill, GitHub notification parsing), incoming chat messages were rejected with "⏳ Busy" because chat replies and background work shared a single worker lane. Split `_run_in_worker` into two concurrent named lanes — `chat` (interactive, notifies when busy) and `bg` (background, silent when busy). Worker skills now dispatch on the `bg` lane so a long background task no longer silences the chat channel. No extra process is forked; both lanes are daemon threads inside the bridge. Closes Anantys-oss#2099 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
When the project being prepared resolves to the same repo as KOAN_ROOT (self-hosting) and it is on a non-base branch, prepare_project_branch now leaves it in place instead of checking out and resetting to the base branch. This lets an operator test a development branch of the launching repo without Koan switching it back to main. All other managed projects still reset to their base branch before each mission.
…omment per review
…ent-2099 feat(bridge): split worker into chat and bg lanes
…aunching-repo-branch fix(git-prep): keep launching repo on its custom branch
…malization Module was already at 100% line/branch coverage. These 8 cases assert distinct input->output behaviors not previously covered: non-string command/context coercion, whitespace/slash-only command normalizing to None, nested-brace context extraction, and unlabeled code-block parsing. Guards the normalization logic against regressions.
Koan-Bot
added a commit
that referenced
this pull request
Jul 11, 2026
Add _flatten_github_alerts() folding `> [!TYPE]` blocks into plain `TYPE: text`. Wire it into _strip_markdown_for_jira() (plan path) and the Jira branches of build_pr_comment_success/_failure (PR path). The GitHub branches keep native alert rendering. Regression tests pin all 5 kinds, the PR-comment Why-line leak, and the pr_submit bypass-path body. Audit of criterion #3: jira_add_comment() is the raw ADF poster, left as-is (flattening it would mangle human /comment blockquotes); all Koan-generated Jira bodies are flattened at the builder layer upstream. Closes Anantys-oss#2305 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
BabyKoan
pushed a commit
that referenced
this pull request
Jul 11, 2026
Add _flatten_github_alerts() folding `> [!TYPE]` blocks into plain `TYPE: text`. Wire it into _strip_markdown_for_jira() (plan path) and the Jira branches of build_pr_comment_success/_failure (PR path). The GitHub branches keep native alert rendering. Regression tests pin all 5 kinds, the PR-comment Why-line leak, and the pr_submit bypass-path body. Audit of criterion #3: jira_add_comment() is the raw ADF poster, left as-is (flattening it would mangle human /comment blockquotes); all Koan-generated Jira bodies are flattened at the builder layer upstream. Closes Anantys-oss#2305 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
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
6 participants
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.
What: Add 8 behavioral edge-case tests for
github_intent._parse_classification.Why:
app/github_intent.pywas already at 100% line/branch coverage, but several distinct input→output behaviors of the classification-normalization path had no asserting test. Those branches were executed by existing tests without their specific outcomes being pinned, leaving the coercion/normalization logic open to silent regression.How: New cases assert: non-string
command/contextcoercion viastr(), whitespace-only and/-only commands normalizing toNone, context whitespace stripping, nested-brace context extraction (rfind must grab the outermost}), unlabeled code-block parsing, and whitespace-only output returningNone.Testing:
pytest koan/tests/test_github_intent.py— 28 passed (was 20), coverage stays 100% line + branch.ruffclean.