Promote dev to main (v0.7.21)

[mabry1985]

Summary

• Promotes dev → main, reconciling 12-commit drift identified in board audit 2026-04-19
• Includes v0.7.20 and v0.7.21 release bumps
• All commits previously reviewed and merged to dev via individual PRs

Commits included

• feat(ava): expand helm toolset, wire GOAP skills, fix ceremony disable bug (feat(ava): expand helm toolset, wire GOAP skills, fix ceremony disable bug #415)
• chore(release): bump to v0.7.21 (chore(release): bump to v0.7.21 #413)
• chore(projects): register protoAgent in projects.yaml (chore(projects): register protoAgent in projects.yaml #414)
• docs(a2a): spell out dual-shape token parsing + INFO logging (docs(a2a): spell out dual-shape token parsing + INFO logging #412)
• feat: upgrade web_search → searxng_search + give Ava fleet health tools (feat: upgrade web_search → searxng_search + Ava fleet health tools #411)
• chore: remove protoaudio + protovoice from agent rollcall (chore: remove protoaudio + protovoice from rollcall #410)
• chore(release): bump to v0.7.20 (chore(release): bump to v0.7.20 #408)
• fix: extension URIs use proto-labs.ai (not protolabs.ai) (fix: extension URIs use proto-labs.ai (not protolabs.ai) #407)
• Back-merge commits (×2) keeping dev in sync with prior main state

Test plan

• CI passes on dev
• No conflicts with main
• Merge using --merge (merge commit — squash would break next promotion DAG)

🤖 Generated with Claude Code

Summary by CodeRabbit

• New Features

  • Enhanced search with category/time-range filters and configurable result limits
  • Ava expanded with new operational skills (CI debugging, incident response, model downshift, issue triage) and richer tools for observation, orchestration, and actions
  • New GitHub issues endpoint and automated Issue Zero goals/actions driving Ava triage

• Improvements

  • PR remediation can self-dispatch; disabled ceremonies now preserved for downstream handling

• Documentation

  • Updated extension URIs and expanded executor/agent runtime and skills docs

• Chores

  • Bumped release to v0.7.21; added protoAgent project; disabled PR audit ceremony

[coderabbitai]

Warning

Rate limit exceeded

@mabry1985 has exceeded the limit for the number of commits that can be reviewed per hour. Please wait 7 minutes and 21 seconds before requesting another review.

Your organization is not enrolled in usage-based pricing. Contact your admin to enable usage-based pricing to continue reviews beyond the rate limit, or try again in 7 minutes and 21 seconds.

⌛ How to resolve this issue?

After the wait time has elapsed, a review can be triggered using the @coderabbitai review command as a PR comment. Alternatively, push new commits to this PR.

We recommend that you space out your commits to avoid hitting the rate limit.

🚦 How do rate limits work?

CodeRabbit enforces hourly rate limits for each developer per organization.

Our paid plans have higher rate limits than the trial, open-source and free plans. In all cases, we re-allow further reviews after a brief timeout.

Please see our FAQ for further information.

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: 5c1e7ee4-0d07-4394-aee6-3b96fcbcda65

📥 Commits

Reviewing files that changed from the base of the PR and between 58daec5 and 5576e72.

📒 Files selected for processing (5)

• lib/plugins/pr-remediator.ts
• src/agent-runtime/tools/bus-tools.ts
• src/api/board.ts
• src/executor/extensions/langfuse-trace.ts
• src/index.ts

📝 Walkthrough

Walkthrough

This PR updates extension URIs (protolabs.ai → proto-labs.ai), replaces web_search with an enhanced searxng_search, adds per-skill systemPromptOverride in DeepAgentExecutor, introduces PR remediator self-dispatch, preserves disabled ceremonies, expands Ava agent tools/skills and Issue Zero goals/actions, and adds GitHub-issues polling/route.

Changes

Cohort / File(s)	Summary
Extension URI Updates docs/extensions/blast-v1.md, docs/extensions/confidence-v1.md, docs/extensions/cost-v1.md, docs/extensions/effect-domain-v1.md, docs/extensions/hitl-mode-v1.md, docs/guides/extend-an-a2a-agent.md, src/executor/extensions/blast.ts, src/executor/extensions/confidence.ts, src/executor/extensions/cost.ts, src/executor/extensions/effect-domain.ts, src/executor/extensions/hitl-mode.ts, src/executor/task-tracker.ts	Replaced https://protolabs.ai/... URIs with https://proto-labs.ai/... in docs and exported constants; task-tracker reference updated.
Search Tool Replacement src/agent-runtime/tools/bus-tools.ts, src/executor/executors/deep-agent-executor.ts	web_search → searxng_search: expanded input schema (category, time_range, max_results), changed request URL construction and headers, increased timeout (10s→15s), and broadened response parsing (results, answers, suggestions, infoboxes).
DeepAgentExecutor prompt handling src/executor/executors/deep-agent-executor.ts, docs/integrations/runtimes/deep-agent.md	Resolve per-skill systemPromptOverride when present (fallback to agent systemPrompt) and append cached world-state snapshot; docs updated accordingly and Ava example adjusted (model, maxTurns, skills, tools).
Ava agent & skills workspace/agents/ava.yaml, docs/reference/agent-skills.md	Expanded Ava tools (observation/orchestration/write/feedback), added GOAP-dispatched skills (debug_ci_failures, fleet_incident_response, downshift_models, investigate_orphaned_skills, issue_triage), removed chain from skill schema and added systemPromptOverride field.
Issue Zero: GitHub issues integration src/api/github.ts, src/index.ts, workspace/actions.yaml, workspace/goals.yaml	New GET /api/github-issues route aggregating open-issue counts (per-repo and totals), registered github_issues polling domain, added Issue Zero goals and actions that trigger Ava triage dispatches and alerts based on thresholds.
PR Remediator / self-dispatch lib/plugins/pr-remediator.ts, src/pr-remediator.test.ts	Added _selfDispatchRemediation() invoked on world.state.updated ticks to drive remediation from cached PR snapshot; broadened AUTO_APPROVE_AUTHORS/title prefixes; tests updated to exercise self-dispatch flows and HITL escalation.
Ceremony loader & tests src/loaders/ceremonyYamlLoader.ts, src/loaders/__tests__/ceremonyYamlLoader.test.ts, src/plugins/__tests__/CeremonyPlugin.test.ts, workspace/ceremonies/board.pr-audit.yaml	Disabled ceremonies preserved (no early-return); tests adjusted; board.pr-audit disabled (enabled: false).
Action dispatch behavior src/plugins/action-dispatcher-plugin.ts	Tier-0 actions now only auto-complete when tier === "tier_0" AND !action.meta.fireAndForget; fireAndForget actions proceed to normal dispatch/publishing.
Tooling / constants updates src/executor/extensions/*.ts, src/executor/task-tracker.ts	Exported extension URI constants updated to new domain; task-tracker emits world-state delta with updated URI reference.
Tests & test adjustments src/pr-remediator.test.ts, src/loaders/__tests__/ceremonyYamlLoader.test.ts, src/plugins/__tests__/CeremonyPlugin.test.ts	Multiple tests updated to reflect self-dispatch, preserved disabled ceremonies, and new control-flow expectations.
Version bumps & project config package.json, dashboard/package.json, workspace/projects.yaml	Package versions updated to 0.7.21; added protoagent project entry (metadata, agents, discord placeholders).
README & docs README.md, docs/explanation/self-improving-loop.md, other docs	README updated to show DeepAgentExecutor and revised executor targets; self-improving loop docs expanded describing chronic-failure → goal proposals, Ava-driven config diffs (propose_config_change), and GOAP-dispatched operational skills.

Estimated code review effort

🎯 4 (Complex) | ⏱️ ~60 minutes

Possibly related issues

• Update extension URIs: protolabs.ai → proto-labs.ai quinn#60: Extension URI constant changes (protolabs.ai → proto-labs.ai) address Quinn agent-card extension discovery mismatch described in the issue.

Possibly related PRs

• feat(goap): issue_zero domain — track open GitHub issues across fleet #419: Overlaps on adding GET /api/github-issues, registering github_issues domain, and Issue Zero goals/actions — strong code-level overlap.
• feat(ava): expand helm toolset, wire GOAP skills, fix ceremony disable bug #415: Overlapping edits around DeepAgentExecutor per-skill prompt handling, ceremony loader behavior, and related agent/docs changes.

Poem

🐰 A hop, a sniff, a searxng query bright,

Ava learns prompts by skill, then sees the night,
Self-dispatching fixes, PRs set right,
Disabled ceremonies sleep till called to light,
Config diffs proposed with carroted delight.

🚥 Pre-merge checks | ✅ 2 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 50.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (2 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title 'Promote dev to main (v0.7.21)' directly and clearly describes the main objective of the PR: promoting the dev branch to main with release version v0.7.21.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch dev

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

Actionable comments posted: 9

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (1)

src/plugins/__tests__/CeremonyPlugin.test.ts (1)

61-79: ⚠️ Potential issue | 🟡 Minor

Test name overstates behavior that isn’t asserted.

At Line 61 the test says disabled ceremonies are “not scheduled,” but assertions at Line 77 and Line 78 only verify load-state (enabled: false). Either assert unscheduled behavior explicitly or rename the test to match what it validates.

✏️ Minimal fix (rename to match current assertions)

-  test("CeremonyPlugin loader: loads disabled ceremonies but does not schedule them", () => {
+  test("CeremonyPlugin loader: loads disabled ceremonies with enabled: false", () => {
@@
-    // Disabled ceremonies are loaded (so hot-reload can cancel timers) but not scheduled
+    // Disabled ceremonies are loaded (so hot-reload can cancel timers)

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@src/plugins/__tests__/CeremonyPlugin.test.ts` around lines 61 - 79, The
test’s title ("CeremonyPlugin loader: loads disabled ceremonies but does not
schedule them") overstates behavior; update the test name to reflect asserted
behavior or add an assertion that checks scheduling. Either rename the test
string to "CeremonyPlugin loader: loads disabled ceremonies" (modify the
test(...) invocation that wraps writeCeremony / plugin.install /
plugin.getCeremonies) or add an explicit assertion after
plugin.install/getCeremonies to verify the ceremony is not scheduled (e.g.,
check the scheduler/timer registry or plugin API that would contain scheduled
jobs) so the title matches the actual checks.

🧹 Nitpick comments (1)

workspace/ceremonies/board.pr-audit.yaml (1)

8-8: Disabling this ceremony will stop scheduled PR audits.
On Line 8, enabled: false correctly unschedules cron execution; just confirm your replacement trigger path is active before merge (to avoid silent audit gaps).

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@workspace/ceremonies/board.pr-audit.yaml` at line 8, You disabled scheduled
PR audits by setting the ceremony flag `enabled: false`; before merging, verify
and activate the replacement trigger path (the new scheduled or event-driven job
that will run PR audits) so audits don't stop silently — confirm the alternative
trigger is deployed, correctly referenced by the PR-audit workflow, and tested
end-to-end to ensure audit runs continue after this change.

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@docs/explanation/self-improving-loop.md`:
- Around line 188-193: The documentation incorrectly states that every
propose_config_change requires an evidence block; update the paragraph so it
clarifies that only tuning-agent proposals require an evidence block returned by
get_confidence_summary (with sampleCount >= 50) or get_cost_summary, while goal
and action proposals do not require evidence. Reference the
propose_config_change tool and explicitly state that evidence is mandatory only
for "tuning-agent" proposals (e.g., agent tuning like "Quinn's pr_review..."),
and remove or reword the blanket sentence that makes evidence sound mandatory
for all config changes.

In `@docs/integrations/runtimes/deep-agent.md`:
- Around line 49-52: The docs list the deprecated tool name "web_search" in the
sample YAML and tools table; update every occurrence of "web_search" to the
runtime's current tool name "searxng_search" (e.g., in the block containing
get_cost_summary, get_confidence_summary, web_search and the same entries at
lines ~114-116) so the example config and tool list match the runtime API and
preserve search functionality when copied.

In `@docs/reference/agent-skills.md`:
- Line 76: The tools list on the line containing get_world_state, get_projects,
get_ci_health, get_pr_pipeline, get_branch_drift, get_outcomes, get_incidents,
get_ceremonies, get_cost_summary, get_confidence_summary currently still lists
web_search; update that entry to searxng_search (and any surrounding text that
references the old tool name) so the documented tool names match the promoted
search-tool migration and avoid confusing agent authors.

In `@lib/plugins/pr-remediator.ts`:
- Around line 361-372: The AUTO_APPROVE_AUTHORS and AUTO_APPROVE_TITLE_PREFIXES
are too permissive (AUTO_APPROVE_AUTHORS, AUTO_APPROVE_TITLE_PREFIXES) and
_runAutoApprovePass() currently auto-submits approvals for any CI-passing match;
tighten this by removing or restricting loose entries like "app/github-actions",
"chore:", and "docs(" from AUTO_APPROVE_TITLE_PREFIXES (or scope them to
specific workflow identities in AUTO_APPROVE_AUTHORS), and add a guarded check
in the auto-approve flow (inside _runAutoApprovePass or a helper it calls) that
verifies changed files are limited to allowed surfaces (e.g., only docs/,
release metadata, dependency manifests, or back-merge-only paths) before
approving; ensure the new verification references the PR file list and uses the
same symbols (AUTO_APPROVE_AUTHORS, AUTO_APPROVE_TITLE_PREFIXES,
_runAutoApprovePass) so the logic only auto-approves when both author/workflow
and restricted file-scope conditions are met.

In `@src/agent-runtime/tools/bus-tools.ts`:
- Around line 401-407: The code uses new URL("/search", endpoint) which discards
any path in SEARXNG_URL; instead, create a URL object from endpoint (const base
= new URL(endpoint)), normalize and append the search path to base.pathname
(e.g. base.pathname = base.pathname.replace(/\/$/, '') + '/search'), then use
that URL (assign to url) so requests go to the configured path-prefixed host;
update references to endpoint and url accordingly.

In `@src/executor/executors/deep-agent-executor.ts`:
- Around line 273-289: The fetch response is parsed without checking HTTP
status, so non-OK SearxNG responses (429/500) will cause resp.json() to throw
and crash the agent; change the code around the fetch/resp handling in
deep-agent-executor.ts to first check resp.ok and, if false, read the response
body (e.g., await resp.text()) and surface a recoverable tool error (match the
behavior in src/agent-runtime/tools/bus-tools.ts by throwing or returning an
error message like `SearxNG returned ${resp.status}: ${body}`) instead of
calling resp.json(); keep the successful path unchanged (use resp.json() and the
existing mapping using input.max_results, cap, and the
results/answers/suggestions/infoboxes transformation).

In `@workspace/agents/ava.yaml`:
- Line 56: The YAML still references the removed tool name web_search in the Ava
prompt text; update those occurrences to the new tool name searxng_search so the
prompt and runtime match. Specifically, replace "web_search" with
"searxng_search" in the Ava system/instruction copy (the lines around where the
prompt lists tools and the line near 190) so the prompt instructs Ava to use the
registered DeepAgentExecutor tool searxng_search and aligns with BUS_TOOL_NAMES
and the registered tool list.
- Around line 114-119: The policy text erroneously requires evidence from
get_cost_summary or get_confidence_summary for every propose_config_change;
change it so only agent-tuning proposals require an evidence block (with
sampleCount >= 50 and data from get_cost_summary or get_confidence_summary),
while new-goal and new-action proposals do not require that evidence; update the
describe/validation text around propose_config_change and references to
get_cost_summary/get_confidence_summary to reflect this distinction.

---

Outside diff comments:
In `@src/plugins/__tests__/CeremonyPlugin.test.ts`:
- Around line 61-79: The test’s title ("CeremonyPlugin loader: loads disabled
ceremonies but does not schedule them") overstates behavior; update the test
name to reflect asserted behavior or add an assertion that checks scheduling.
Either rename the test string to "CeremonyPlugin loader: loads disabled
ceremonies" (modify the test(...) invocation that wraps writeCeremony /
plugin.install / plugin.getCeremonies) or add an explicit assertion after
plugin.install/getCeremonies to verify the ceremony is not scheduled (e.g.,
check the scheduler/timer registry or plugin API that would contain scheduled
jobs) so the title matches the actual checks.

---

Nitpick comments:
In `@workspace/ceremonies/board.pr-audit.yaml`:
- Line 8: You disabled scheduled PR audits by setting the ceremony flag
`enabled: false`; before merging, verify and activate the replacement trigger
path (the new scheduled or event-driven job that will run PR audits) so audits
don't stop silently — confirm the alternative trigger is deployed, correctly
referenced by the PR-audit workflow, and tested end-to-end to ensure audit runs
continue after this change.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: e152dea7-e156-4e25-8c8c-e5b9a96a4d10

📥 Commits

Reviewing files that changed from the base of the PR and between fc4ce52 and 9dc4b79.

📒 Files selected for processing (29)

• README.md
• dashboard/package.json
• docs/explanation/self-improving-loop.md
• docs/extensions/blast-v1.md
• docs/extensions/confidence-v1.md
• docs/extensions/cost-v1.md
• docs/extensions/effect-domain-v1.md
• docs/extensions/hitl-mode-v1.md
• docs/guides/extend-an-a2a-agent.md
• docs/integrations/runtimes/deep-agent.md
• docs/reference/agent-skills.md
• lib/plugins/pr-remediator.ts
• package.json
• src/agent-runtime/tools/bus-tools.ts
• src/executor/executors/deep-agent-executor.ts
• src/executor/extensions/blast.ts
• src/executor/extensions/confidence.ts
• src/executor/extensions/cost.ts
• src/executor/extensions/effect-domain.ts
• src/executor/extensions/hitl-mode.ts
• src/executor/task-tracker.ts
• src/loaders/__tests__/ceremonyYamlLoader.test.ts
• src/loaders/ceremonyYamlLoader.ts
• src/plugins/__tests__/CeremonyPlugin.test.ts
• src/plugins/action-dispatcher-plugin.ts
• src/pr-remediator.test.ts
• workspace/agents/ava.yaml
• workspace/ceremonies/board.pr-audit.yaml
• workspace/projects.yaml

[coderabbitai]

⚠️ Potential issue | 🟡 Minor

Don't document evidence as mandatory for every config change.

propose_config_change only requires an evidence block for tuning-agent proposals. As written, Line 193 contradicts the tool contract and makes goal/action proposals sound stricter than they are.

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@docs/explanation/self-improving-loop.md` around lines 188 - 193, The
documentation incorrectly states that every propose_config_change requires an
evidence block; update the paragraph so it clarifies that only tuning-agent
proposals require an evidence block returned by get_confidence_summary (with
sampleCount >= 50) or get_cost_summary, while goal and action proposals do not
require evidence. Reference the propose_config_change tool and explicitly state
that evidence is mandatory only for "tuning-agent" proposals (e.g., agent tuning
like "Quinn's pr_review..."), and remove or reword the blanket sentence that
makes evidence sound mandatory for all config changes.

[coderabbitai]

⚠️ Potential issue | 🟡 Minor

Replace web_search with searxng_search in the docs.

The runtime now exposes searxng_search; keeping web_search in the sample YAML and tool table documents a tool name that no longer exists. Copy-pasted configs from this page will silently lose search support.

Also applies to: 114-116

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@docs/integrations/runtimes/deep-agent.md` around lines 49 - 52, The docs list
the deprecated tool name "web_search" in the sample YAML and tools table; update
every occurrence of "web_search" to the runtime's current tool name
"searxng_search" (e.g., in the block containing get_cost_summary,
get_confidence_summary, web_search and the same entries at lines ~114-116) so
the example config and tool list match the runtime API and preserve search
functionality when copied.

[coderabbitai]

⚠️ Potential issue | 🟡 Minor

Tool list appears stale: web_search should likely be searxng_search.

Line 76 still documents web_search, which conflicts with this promotion’s stated search-tool migration and can mislead agent authors.

🛠️ Suggested doc fix

-**Observation:** `get_world_state`, `get_projects`, `get_ci_health`, `get_pr_pipeline`, `get_branch_drift`, `get_outcomes`, `get_incidents`, `get_ceremonies`, `get_cost_summary`, `get_confidence_summary`, `web_search`
+**Observation:** `get_world_state`, `get_projects`, `get_ci_health`, `get_pr_pipeline`, `get_branch_drift`, `get_outcomes`, `get_incidents`, `get_ceremonies`, `get_cost_summary`, `get_confidence_summary`, `searxng_search`

📝 Committable suggestion

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

Suggested change

	**Observation:** `get_world_state`, `get_projects`, `get_ci_health`, `get_pr_pipeline`, `get_branch_drift`, `get_outcomes`, `get_incidents`, `get_ceremonies`, `get_cost_summary`, `get_confidence_summary`, `web_search`
	**Observation:** `get_world_state`, `get_projects`, `get_ci_health`, `get_pr_pipeline`, `get_branch_drift`, `get_outcomes`, `get_incidents`, `get_ceremonies`, `get_cost_summary`, `get_confidence_summary`, `searxng_search`

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@docs/reference/agent-skills.md` at line 76, The tools list on the line
containing get_world_state, get_projects, get_ci_health, get_pr_pipeline,
get_branch_drift, get_outcomes, get_incidents, get_ceremonies, get_cost_summary,
get_confidence_summary currently still lists web_search; update that entry to
searxng_search (and any surrounding text that references the old tool name) so
the documented tool names match the promoted search-tool migration and avoid
confusing agent authors.

[coderabbitai]

⚠️ Potential issue | 🟠 Major

Narrow the new auto-approve classes.

app/github-actions, chore:, and docs( are not strong safety signals on their own. With _runAutoApprovePass() later auto-submitting APPROVED reviews for any passing-CI match, this can remove the last human gate from arbitrary code-changing PRs that happen to use those prefixes. Limit this to tightly controlled identities/workflows, or verify the changed files are actually restricted to docs/release/back-merge surfaces before approving.

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@lib/plugins/pr-remediator.ts` around lines 361 - 372, The
AUTO_APPROVE_AUTHORS and AUTO_APPROVE_TITLE_PREFIXES are too permissive
(AUTO_APPROVE_AUTHORS, AUTO_APPROVE_TITLE_PREFIXES) and _runAutoApprovePass()
currently auto-submits approvals for any CI-passing match; tighten this by
removing or restricting loose entries like "app/github-actions", "chore:", and
"docs(" from AUTO_APPROVE_TITLE_PREFIXES (or scope them to specific workflow
identities in AUTO_APPROVE_AUTHORS), and add a guarded check in the auto-approve
flow (inside _runAutoApprovePass or a helper it calls) that verifies changed
files are limited to allowed surfaces (e.g., only docs/, release metadata,
dependency manifests, or back-merge-only paths) before approving; ensure the new
verification references the PR file list and uses the same symbols
(AUTO_APPROVE_AUTHORS, AUTO_APPROVE_TITLE_PREFIXES, _runAutoApprovePass) so the
logic only auto-approves when both author/workflow and restricted file-scope
conditions are met.

[coderabbitai]

⚠️ Potential issue | 🟡 Minor

Enforce the documented default category and validate max_results.

The schema says category defaults to general, but omitted requests never set categories, so behavior depends on server config instead of this tool contract. max_results also accepts negative or fractional values, which feed straight into slice(0, cap).

Suggested fix

-      category: z.enum(["general", "news", "science", "it"]).optional().describe("SearXNG category. Defaults to general."),
-      max_results: z.number().optional().describe("Max results to return. Default: 10."),
+      category: z.enum(["general", "news", "science", "it"]).default("general").describe("SearXNG category."),
+      max_results: z.number().int().positive().optional().describe("Max results to return. Default: 10."),
     },
-    async ({ query, category, time_range, max_results }) => {
+    async ({ query, category = "general", time_range, max_results }) => {
       const endpoint = process.env.SEARXNG_URL ?? "http://searxng:8080";
       try {
         const url = new URL("/search", endpoint);
         url.searchParams.set("q", query);
         url.searchParams.set("format", "json");
-        if (category) url.searchParams.set("categories", category);
+        url.searchParams.set("categories", category);

[coderabbitai]

⚠️ Potential issue | 🟠 Major

Preserve path-prefixed SEARXNG_URL values.

new URL("/search", endpoint) discards any path component in SEARXNG_URL. If this is configured as something like https://search.example.com/searxng, requests will go to /search at the origin root instead of /searxng/search.

Suggested fix

-        const url = new URL("/search", endpoint);
+        const base = endpoint.endsWith("/") ? endpoint : `${endpoint}/`;
+        const url = new URL("search", base);

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@src/agent-runtime/tools/bus-tools.ts` around lines 401 - 407, The code uses
new URL("/search", endpoint) which discards any path in SEARXNG_URL; instead,
create a URL object from endpoint (const base = new URL(endpoint)), normalize
and append the search path to base.pathname (e.g. base.pathname =
base.pathname.replace(/\/$/, '') + '/search'), then use that URL (assign to url)
so requests go to the configured path-prefixed host; update references to
endpoint and url accordingly.

[coderabbitai]

⚠️ Potential issue | 🟠 Major

Handle non-OK SearXNG responses before parsing JSON.

A 429/500 response currently falls straight into resp.json(), so an upstream error can throw out of the tool and fail the whole agent run. src/agent-runtime/tools/bus-tools.ts:387-435 already treats non-OK SearXNG responses as a recoverable tool failure; this copy should match it.

Suggested fix

         const resp = await fetch(url.toString(), {
           headers: { Accept: "application/json" },
           signal: AbortSignal.timeout(15_000),
         });
+        if (!resp.ok) {
+          return JSON.stringify({ results: [], error: `SearXNG ${resp.status}` });
+        }
         const data = await resp.json() as {
           results?: Array<{ title: string; url: string; content: string; engine: string; score?: number }>;
           answers?: string[];
           suggestions?: string[];
           infoboxes?: Array<{ infobox: string; content: string }>;

📝 Committable suggestion

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

Suggested change

	const resp = await fetch(url.toString(), {
	headers: { Accept: "application/json" },
	signal: AbortSignal.timeout(15_000),
	});
	const data = await resp.json() as {
	results?: Array<{ title: string; url: string; content: string; engine: string; score?: number }>;
	answers?: string[];
	suggestions?: string[];
	infoboxes?: Array<{ infobox: string; content: string }>;
	};
	const cap = input.max_results ?? 10;
	return JSON.stringify({
	results: (data.results ?? []).slice(0, cap).map(r => ({ title: r.title, url: r.url, snippet: r.content, engine: r.engine })),
	...(data.answers?.length ? { answers: data.answers } : {}),
	...(data.suggestions?.length ? { suggestions: data.suggestions } : {}),
	...(data.infoboxes?.length ? { infoboxes: data.infoboxes.map(ib => ({ title: ib.infobox, content: ib.content })) } : {}),
	});
	const resp = await fetch(url.toString(), {
	headers: { Accept: "application/json" },
	signal: AbortSignal.timeout(15_000),
	});
	if (!resp.ok) {
	return JSON.stringify({ results: [], error: `SearXNG ${resp.status}` });
	}
	const data = await resp.json() as {
	results?: Array<{ title: string; url: string; content: string; engine: string; score?: number }>;
	answers?: string[];
	suggestions?: string[];
	infoboxes?: Array<{ infobox: string; content: string }>;
	};
	const cap = input.max_results ?? 10;
	return JSON.stringify({
	results: (data.results ?? []).slice(0, cap).map(r => ({ title: r.title, url: r.url, snippet: r.content, engine: r.engine })),
	...(data.answers?.length ? { answers: data.answers } : {}),
	...(data.suggestions?.length ? { suggestions: data.suggestions } : {}),
	...(data.infoboxes?.length ? { infoboxes: data.infoboxes.map(ib => ({ title: ib.infobox, content: ib.content })) } : {}),
	});

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@src/executor/executors/deep-agent-executor.ts` around lines 273 - 289, The
fetch response is parsed without checking HTTP status, so non-OK SearxNG
responses (429/500) will cause resp.json() to throw and crash the agent; change
the code around the fetch/resp handling in deep-agent-executor.ts to first check
resp.ok and, if false, read the response body (e.g., await resp.text()) and
surface a recoverable tool error (match the behavior in
src/agent-runtime/tools/bus-tools.ts by throwing or returning an error message
like `SearxNG returned ${resp.status}: ${body}`) instead of calling resp.json();
keep the successful path unchanged (use resp.json() and the existing mapping
using input.max_results, cap, and the results/answers/suggestions/infoboxes
transformation).

[coderabbitai]

⚠️ Potential issue | 🟠 Major

Rename the removed web_search tool here.

DeepAgentExecutor now registers searxng_search, and BUS_TOOL_NAMES no longer includes web_search. Keeping the old name in Lines 56 and 190 means Ava loses web search at runtime while the prompt still tells her to use it.

Suggested fix

-  - **web_search** — Quick web search. For deep research, delegate to Researcher.
+  - **searxng_search** — Quick web search. For deep research, delegate to Researcher.
...
-  - web_search
+  - searxng_search

Also applies to: 190-190

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@workspace/agents/ava.yaml` at line 56, The YAML still references the removed
tool name web_search in the Ava prompt text; update those occurrences to the new
tool name searxng_search so the prompt and runtime match. Specifically, replace
"web_search" with "searxng_search" in the Ava system/instruction copy (the lines
around where the prompt lists tools and the line near 190) so the prompt
instructs Ava to use the registered DeepAgentExecutor tool searxng_search and
aligns with BUS_TOOL_NAMES and the registered tool list.

[coderabbitai]

⚠️ Potential issue | 🟠 Major

Don't require cost/confidence evidence for every config proposal.

propose_config_change only requires an evidence block for agent-tuning proposals. Telling Ava that all proposals must come from get_cost_summary or get_confidence_summary will block the new-goal/new-action paths this PR is adding.

Suggested fix

-  All proposals require evidence from `get_cost_summary` or
-  `get_confidence_summary`. Include sampleCount >= 50 for agent tuning.
+  Data-driven tuning proposals should include evidence from
+  `get_cost_summary` or `get_confidence_summary`.
+  Include sampleCount >= 50 for agent tuning.
+  Goal/action proposals may instead rely on the triggering incident,
+  outcome cluster, or world-state evidence.

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@workspace/agents/ava.yaml` around lines 114 - 119, The policy text
erroneously requires evidence from get_cost_summary or get_confidence_summary
for every propose_config_change; change it so only agent-tuning proposals
require an evidence block (with sampleCount >= 50 and data from get_cost_summary
or get_confidence_summary), while new-goal and new-action proposals do not
require that evidence; update the describe/validation text around
propose_config_change and references to get_cost_summary/get_confidence_summary
to reflect this distinction.

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@src/api/github.ts`:
- Around line 489-490: The ghApi call that fetches issues using
`/repos/${repo}/issues?state=open&per_page=100&sort=created&direction=asc` only
retrieves the first page; change the logic in the function containing that ghApi
invocation to implement pagination similar to handleGetPrPipeline(): loop up to
5 pages, call ghApi with a page parameter (e.g., `&page=${page}`), accumulate
results into the existing issues array (or a new aggregated array), break early
if a page returns an empty array or fewer than 100 items, and then use the
aggregated list for downstream counts/processing; ensure you reference and
update the same variable names used around the original ghApi call so other code
continues to work.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: 2d779bc6-9e8d-4cd5-8f6f-77d6c2012eae

📥 Commits

Reviewing files that changed from the base of the PR and between 9dc4b79 and 58daec5.

📒 Files selected for processing (5)

• src/api/github.ts
• src/index.ts
• workspace/actions.yaml
• workspace/agents/ava.yaml
• workspace/goals.yaml

✅ Files skipped from review due to trivial changes (1)

• src/index.ts

[coderabbitai]

⚠️ Potential issue | 🟡 Minor

No pagination — repos with >100 open issues will be undercounted.

The query fetches only the first 100 issues (per_page=100) without a pagination loop. Compare with handleGetPrPipeline() (lines 142-157) which iterates up to 5 pages. While the "Issue Zero" goal targets <5 total issues, accurate counts matter for triage visibility.

♻️ Suggested pagination pattern (mirroring PR pipeline)

-        const issues = await ghApi(
-          `/repos/${repo}/issues?state=open&per_page=100&sort=created&direction=asc`,
-        ) as Array<{
+        const allIssues: Array<{
           number: number;
           title: string;
           labels: Array<{ name: string }>;
           created_at: string;
           updated_at: string;
           user: { login: string } | null;
           pull_request?: unknown;
-        }>;
+        }> = [];
+        const MAX_PAGES = 5;
+        for (let page = 1; page <= MAX_PAGES; page++) {
+          const batch = await ghApi(
+            `/repos/${repo}/issues?state=open&per_page=100&page=${page}&sort=created&direction=asc`,
+          ) as typeof allIssues;
+          if (batch.length === 0) break;
+          allIssues.push(...batch);
+          if (batch.length < 100) break;
+        }
+        const issues = allIssues;

📝 Committable suggestion

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

Suggested change

	const issues = await ghApi(
	`/repos/${repo}/issues?state=open&per_page=100&sort=created&direction=asc`,
	const allIssues: Array<{
	number: number;
	title: string;
	labels: Array<{ name: string }>;
	created_at: string;
	updated_at: string;
	user: { login: string } | null;
	pull_request?: unknown;
	}> = [];
	const MAX_PAGES = 5;
	for (let page = 1; page <= MAX_PAGES; page++) {
	const batch = await ghApi(
	`/repos/${repo}/issues?state=open&per_page=100&page=${page}&sort=created&direction=asc`,
	) as typeof allIssues;
	if (batch.length === 0) break;
	allIssues.push(...batch);
	if (batch.length < 100) break;
	}
	const issues = allIssues;

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@src/api/github.ts` around lines 489 - 490, The ghApi call that fetches issues
using `/repos/${repo}/issues?state=open&per_page=100&sort=created&direction=asc`
only retrieves the first page; change the logic in the function containing that
ghApi invocation to implement pagination similar to handleGetPrPipeline(): loop
up to 5 pages, call ghApi with a page parameter (e.g., `&page=${page}`),
accumulate results into the existing issues array (or a new aggregated array),
break early if a page returns an empty array or fewer than 100 items, and then
use the aggregated list for downstream counts/processing; ensure you reference
and update the same variable names used around the original ghApi call so other
code continues to work.