feat(docs-tools): fan out subagents per requirement#135
Merged
Conversation
|
Note Reviews pausedIt looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the Use the following commands to manage reviews:
Use the checkboxes below for quick actions:
WalkthroughAdds a multi-agent fan-out architecture to docs-tools: two new agents ( Changes
Sequence Diagram(s)sequenceDiagram
participant Orchestrator as Orchestrator
participant Discover as requirements-discoverer
participant Analyst as requirements-analyst
participant Classifier as evidence-classifier
participant CodeFinder as Code-Finder
participant JIRA as JIRA/Git/Docs
Orchestrator->>Discover: request discovery (ticket key / context)
Discover->>JIRA: fetch ticket + graph traversal
JIRA-->>Discover: discovered requirements + PR/Doc URLs
Orchestrator->>Analyst: dispatch per-requirement (parallel)
Orchestrator->>Classifier: dispatch per-requirement (parallel)
Analyst->>JIRA: fetch referenced docs / PR diffs / web fetch
Classifier->>CodeFinder: run code-finder search (repo/query/limit)
CodeFinder-->>Classifier: ranked results + scores
Analyst-->>Orchestrator: JSON result (per requirement)
Classifier-->>Orchestrator: JSON result (status: grounded/partial/absent)
Orchestrator->>Orchestrator: merge results, apply fallbacks, write `requirements.md` + `step-result.json`
Estimated code review effort🎯 4 (Complex) | ⏱️ ~45 minutes Possibly related PRs
Important Pre-merge checks failedPlease resolve all errors before merging. Addressing warnings is optional. ❌ Failed checks (1 error)
✅ Passed checks (9 passed)
✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 2
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (2)
plugins/docs-tools/skills/docs-workflow-scope-req-audit/SKILL.md (1)
111-124:⚠️ Potential issue | 🟠 Major | ⚡ Quick winMake requirement parsing resilient to the actual
requirements.mdshape.The parser contract here is too strict and no longer matches the format documented in
plugins/docs-tools/skills/docs-workflow-requirements/SKILL.md(REQ headings shown as#### REQ-NNNand summary as- **Summary**: ...). This can trigger false “No requirements found” failures.Suggested parser-contract update
-Read `$REQUIREMENTS_FILE` and extract each requirement. The requirements-analyst produces requirements in this pattern: +Read `$REQUIREMENTS_FILE` and extract each requirement using a heading-level-agnostic pattern. The assembled requirements.md contains entries in this shape: ```text -### REQ-NNN: [title] +#### REQ-NNN: [title] -**Summary**: [description] +- **Summary**: [description]For each requirement, extract:
id— the REQ-NNN identifiertitle— the requirement titlesummary— the summary text-If no requirements are found matching this pattern, STOP with error: "No requirements found in requirements.md. Expected REQ-NNN pattern."
+If no requirements are found matching^#{3,6}\s+REQ-\d+:\s+, STOP with error: "No requirements found in requirements.md. Expected REQ-NNN heading pattern."</details> <details> <summary>🤖 Prompt for AI Agents</summary>Verify each finding against the current code and only fix it if needed.
In
@plugins/docs-tools/skills/docs-workflow-scope-req-audit/SKILL.mdaround
lines 111 - 124, Update the requirements parser to be resilient to heading
levels and minor formatting differences: change the strict pattern that looks
for "### REQ-NNN: [title]" and "Summary: [description]" (referenced near
$REQUIREMENTS_FILE) to a more flexible regex that matches headings with 3–6
hashes and numeric IDs (for example use pattern '^#{3,6}\s+REQ-\d+:\s+' to
detect requirement headings) and accept an optional space before "Summary"
(e.g., '- Summary: ...'); also update the STOP error string from "No
requirements found in requirements.md. Expected REQ-NNN pattern." to "No
requirements found in requirements.md. Expected REQ-NNN heading pattern." so
failures reflect the relaxed match.</details> </blockquote></details> <details> <summary>plugins/docs-tools/skills/docs-workflow-requirements/SKILL.md (1)</summary><blockquote> `33-271`: _🛠️ Refactor suggestion_ | _🟠 Major_ | _🏗️ Heavy lift_ **Move procedural orchestration from SKILL.md into `scripts/` for this step.** This file now embeds substantial operational logic (arg parsing, validation, dispatch, merge, output assembly). Please shift executable procedure into `skills/docs-workflow-requirements/scripts/` and keep SKILL.md as dispatch + contract + verification only. As per coding guidelines: "Skill logic must defer procedural logic (argument parsing, mode determination, input validation, path computation, directory creation) to a script under skills/<skill-name>/scripts/. The SKILL.md should only contain frontmatter, instructions, domain knowledge, dispatch instructions, and verification." <details> <summary>🤖 Prompt for AI Agents</summary>Verify each finding against the current code and only fix it if needed.
In
@plugins/docs-tools/skills/docs-workflow-requirements/SKILL.mdaround lines
33 - 271, SKILL.md currently contains procedural orchestration (argument
parsing, validation, dispatch, merging, and file I/O) that should be moved into
an executable script; extract all operational logic from SKILL.md and implement
it as a script under skills/docs-workflow-requirements/scripts/ (e.g., a single
entrypoint script that performs: parse args (--base-path, --pr, ticket id),
create OUTPUT_DIR, run the discovery agent and validate DISCOVERY_FILE, fan out
requirements-analyst calls in one message, parse/merge agent results into the
required merged structure, assemble OUTPUT_FILE (requirements.md) using the
exact document contract, run parse_title.py and write step-result.json, and
verify outputs); leave SKILL.md only with dispatch instructions, agent prompts,
input/output contract, and verification steps, and update SKILL.md to reference
the new script for the procedural execution.</details> </blockquote></details> </blockquote></details>🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed. Inline comments: In `@plugins/docs-tools/skills/docs-workflow-requirements/SKILL.md`: - Line 50: Two fenced code blocks containing the YAML snippets that start with "Agent:" and include "subagent_type: docs-tools:requirements-discoverer" and "subagent_type: docs-tools:requirements-analyst" are currently unlabeled and trigger MD040; update each opening fence from ``` to ```yaml so both blocks are explicitly marked as YAML (leave the inner content and closing fences unchanged). In `@plugins/docs-tools/skills/docs-workflow-scope-req-audit/SKILL.md`: - Line 158: The markdown fenced code block around the "Agent:" YAML example in SKILL.md is missing a language tag (triggering MD040); update the opening fence to ```yaml so the block is explicitly YAML (e.g., change ``` to ```yaml for the block containing "Agent:", "subagent_type", and "model") to satisfy the linter. --- Outside diff comments: In `@plugins/docs-tools/skills/docs-workflow-requirements/SKILL.md`: - Around line 33-271: SKILL.md currently contains procedural orchestration (argument parsing, validation, dispatch, merging, and file I/O) that should be moved into an executable script; extract all operational logic from SKILL.md and implement it as a script under skills/docs-workflow-requirements/scripts/ (e.g., a single entrypoint script that performs: parse args (--base-path, --pr, ticket id), create OUTPUT_DIR, run the discovery agent and validate DISCOVERY_FILE, fan out requirements-analyst calls in one message, parse/merge agent results into the required merged structure, assemble OUTPUT_FILE (requirements.md) using the exact document contract, run parse_title.py and write step-result.json, and verify outputs); leave SKILL.md only with dispatch instructions, agent prompts, input/output contract, and verification steps, and update SKILL.md to reference the new script for the procedural execution. In `@plugins/docs-tools/skills/docs-workflow-scope-req-audit/SKILL.md`: - Around line 111-124: Update the requirements parser to be resilient to heading levels and minor formatting differences: change the strict pattern that looks for "### REQ-NNN: [title]" and "**Summary**: [description]" (referenced near $REQUIREMENTS_FILE) to a more flexible regex that matches headings with 3–6 hashes and numeric IDs (for example use pattern '^#{3,6}\s+REQ-\d+:\s+' to detect requirement headings) and accept an optional space before "**Summary**" (e.g., '- **Summary**: ...'); also update the STOP error string from "No requirements found in requirements.md. Expected REQ-NNN pattern." to "No requirements found in requirements.md. Expected REQ-NNN heading pattern." so failures reflect the relaxed match.🪄 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: Enterprise
Run ID:
e4d6d6cd-d61d-4b24-9629-3237f5b69138📒 Files selected for processing (7)
plugins/docs-tools/.claude-plugin/plugin.jsonplugins/docs-tools/agents/evidence-classifier.mdplugins/docs-tools/agents/requirements-analyst.mdplugins/docs-tools/agents/requirements-discoverer.mdplugins/docs-tools/skills/docs-workflow-requirements/SKILL.mdplugins/docs-tools/skills/docs-workflow-scope-req-audit/SKILL.mdplugins/docs-tools/skills/docs-workflow-scope-req-audit/prompts/gap-classification.md💤 Files with no reviewable changes (1)
- plugins/docs-tools/skills/docs-workflow-scope-req-audit/prompts/gap-classification.md
There was a problem hiding this comment.
Actionable comments posted: 1
🧹 Nitpick comments (3)
plugins/docs-tools/agents/requirements-analyst.md (1)
5-5: ⚡ Quick winStandardize skill references to fully qualified names in this agent doc.
This file uses bare skill names in frontmatter and inline instructions (
jira-reader,git-pr-reader,docs-convert-gdoc-md). Please align to one canonical convention for agents and use fully qualifiedplugin:skillnames consistently to avoid ambiguous invocation behavior.Suggested edit
-skills: jira-reader, git-pr-reader, article-extractor, redhat-docs-toc, docs-convert-gdoc-md +skills: docs-tools:jira-reader, docs-tools:git-pr-reader, docs-tools:article-extractor, docs-tools:redhat-docs-toc, docs-tools:docs-convert-gdoc-md-Use the `docs-convert-gdoc-md` skill for Google Docs. For other specs, use WebFetch. +Use the `docs-tools:docs-convert-gdoc-md` skill for Google Docs. For other specs, use WebFetch.Based on learnings: “Always use fully qualified
plugin:skillnames … in agent instructions, Skill invocations, inline text references, and cross-references between skills.”Also applies to: 56-56
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@plugins/docs-tools/agents/requirements-analyst.md` at line 5, Frontmatter and inline references use bare skill names (jira-reader, git-pr-reader, article-extractor, redhat-docs-toc, docs-convert-gdoc-md) which must be replaced with fully qualified plugin:skill names; update the frontmatter skills line and every inline mention (including the occurrence around line 56) to the canonical form (e.g., plugin-name:jira-reader, plugin-name:git-pr-reader, plugin-name:article-extractor, plugin-name:redhat-docs-toc, plugin-name:docs-convert-gdoc-md) so all agent instructions, Skill invocations, and cross-references consistently use the fully qualified names.plugins/docs-tools/agents/evidence-classifier.md (2)
114-116: 💤 Low valueMinor: Clarify key_files ordering after deduplication.
Line 116 specifies "file paths from the top 3 results (deduplicated)" but doesn't indicate whether the deduplicated list should preserve rank order. Consider adding brief clarification:
-- `key_files`: file paths from the top 3 results (deduplicated) +- `key_files`: file paths from the top 3 results (deduplicated, preserving rank order)🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@plugins/docs-tools/agents/evidence-classifier.md` around lines 114 - 116, Clarify that the "key_files" field should preserve the original ranking: after deduplicating file paths from the top results, keep them in the same order they appeared in the ranked results and then take up to the first three unique paths (i.e., deduplicate while preserving rank, not sorting alphabetically), so update the description for the key_files definition accordingly.
10-12: 💤 Low valueConsider rewording to reduce repetition.
Three consecutive sentences begin with "You," which affects readability. Consider restructuring:
♻️ Suggested refactor
-You are a code evidence classifier. You receive a single documentation requirement and determine whether the described feature is implemented in the source repository by running a code search and applying score thresholds. +You are a code evidence classifier. Receive a single documentation requirement and determine whether the described feature is implemented in the source repository by running a code search and applying score thresholds. -You produce exactly one JSON object on stdout — no markdown, no commentary, no explanation. +Produce exactly one JSON object on stdout — no markdown, no commentary, no explanation.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@plugins/docs-tools/agents/evidence-classifier.md` around lines 10 - 12, The three opening sentences ("You are a code evidence classifier.", "You receive a single documentation requirement...", "You produce exactly one JSON object...") are repetitive; rewrite them into a single, concise lead that combines role, input, and output constraints (for example: "As a code evidence classifier, take one documentation requirement as input, search the repo, and output exactly one JSON object on stdout — no markdown or commentary."). Update the text containing the exact phrases "You are a code evidence classifier.", "You receive a single documentation requirement", and "You produce exactly one JSON object on stdout" to the new combined sentence and ensure punctuation/constraints are preserved.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@plugins/docs-tools/agents/evidence-classifier.md`:
- Around line 76-80: Clarify the ambiguous classification rules in
evidence-classifier.md by explicitly defining precedence and the edge case:
state that Grounded is evaluated first (requires top hit scores.combined >=
grounded threshold AND at least 2 results with scores > absent threshold), and
if the top hit >= grounded but fewer than 2 results exceed the absent threshold
classify as Partial; also reword the Partial rule to explicitly cover "top hit
between absent and grounded OR top hit >= grounded but fewer than 2 results
above absent" and keep Absent as currently written.
---
Nitpick comments:
In `@plugins/docs-tools/agents/evidence-classifier.md`:
- Around line 114-116: Clarify that the "key_files" field should preserve the
original ranking: after deduplicating file paths from the top results, keep them
in the same order they appeared in the ranked results and then take up to the
first three unique paths (i.e., deduplicate while preserving rank, not sorting
alphabetically), so update the description for the key_files definition
accordingly.
- Around line 10-12: The three opening sentences ("You are a code evidence
classifier.", "You receive a single documentation requirement...", "You produce
exactly one JSON object...") are repetitive; rewrite them into a single, concise
lead that combines role, input, and output constraints (for example: "As a code
evidence classifier, take one documentation requirement as input, search the
repo, and output exactly one JSON object on stdout — no markdown or
commentary."). Update the text containing the exact phrases "You are a code
evidence classifier.", "You receive a single documentation requirement", and
"You produce exactly one JSON object on stdout" to the new combined sentence and
ensure punctuation/constraints are preserved.
In `@plugins/docs-tools/agents/requirements-analyst.md`:
- Line 5: Frontmatter and inline references use bare skill names (jira-reader,
git-pr-reader, article-extractor, redhat-docs-toc, docs-convert-gdoc-md) which
must be replaced with fully qualified plugin:skill names; update the frontmatter
skills line and every inline mention (including the occurrence around line 56)
to the canonical form (e.g., plugin-name:jira-reader, plugin-name:git-pr-reader,
plugin-name:article-extractor, plugin-name:redhat-docs-toc,
plugin-name:docs-convert-gdoc-md) so all agent instructions, Skill invocations,
and cross-references consistently use the fully qualified names.
🪄 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: Enterprise
Run ID: 9986819c-5007-4478-b3ad-df3a3df966f7
📒 Files selected for processing (2)
plugins/docs-tools/agents/evidence-classifier.mdplugins/docs-tools/agents/requirements-analyst.md
There was a problem hiding this comment.
Actionable comments posted: 3
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@plugins/docs-tools/agents/requirements-analyst.md`:
- Around line 22-25: The hard-fail rule that demands successful access to
JIRA/Git should be made source-aware: change the logic that unconditionally
requires JIRA/Git access to first inspect the requirement object’s sources list
and only attempt/require access for systems actually listed (i.e., check for
"jira" and/or "git" in requirement.sources); if a listed source fails access,
then return the error result, otherwise do not fail for missing/unlisted
systems. Apply the same conditional-source check to the other occurrences of
this rule in the file (the block referenced around lines 39-64) so spec-only or
JIRA-only requirements aren’t erroneously hard-failed.
In `@plugins/docs-tools/agents/requirements-discoverer.md`:
- Around line 99-104: The related_tickets contract is inconsistent between
per-ticket "relation" and containerized groups (parent, children, siblings,
linked); fix by normalizing to a single representation: when assembling
related_tickets ensure every ticket object includes the fields key, url, summary
and a relation string (e.g., "parent", "child", "sibling", "linked") and stop
relying on container-only groupings, or if you keep containers, populate the
relation field for each ticket within the containers (update the code that
builds related_tickets to set ticket.relation accordingly and remove ambiguity
between container names and per-ticket relation).
- Around line 22-25: The current access gate in requirements-discoverer mandates
successful Git/JIRA access always; change it so Git access is only required when
a PR/MR URL is present in inputs or auto-discovery (detect via the PR/MR URL
field), while JIRA/spec-derived discovery proceeds even if Git is unavailable;
when skipping Git due to missing PR/MR, do not set the JSON "error" field for
Git failures—only set "error" when a required resource (e.g., a referenced PR/MR
or mandatory JIRA ticket) actually fails to be fetched.
🪄 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: Enterprise
Run ID: 42d138af-9929-4e91-b05c-e946f5e4460c
📒 Files selected for processing (2)
plugins/docs-tools/agents/requirements-analyst.mdplugins/docs-tools/agents/requirements-discoverer.md
There was a problem hiding this comment.
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 `@plugins/docs-tools/agents/evidence-classifier.md`:
- Line 5: The configuration in evidence-classifier (the maxTurns setting) is
inconsistent with the PR summary: change the manifest value for maxTurns from 8
to 5 (or update the PR summary if 8 is intended); locate the maxTurns entry in
the evidence-classifier agent definition (the maxTurns key in the
evidence-classifier.md file) and set it to the correct integer (5) so the agent
configuration matches the intended design.
🪄 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: Enterprise
Run ID: a557b92e-6dae-43b6-b3e0-041bd0149da7
📒 Files selected for processing (1)
plugins/docs-tools/agents/evidence-classifier.md
There was a problem hiding this comment.
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 `@plugins/docs-tools/skills/docs-workflow-scope-req-audit/SKILL.md`:
- Around line 180-197: The doc now describes a fallback requirement entry that
uses status: "error", which conflicts with the claimed unchanged contract of
only "grounded|partial|absent"; either update the contract to explicitly allow
"error" or change the fallback to preserve the existing contract (set "status":
"absent" and add an optional "error" field for diagnostics), and update the
example JSON and the text that says "unchanged output format" so downstream
parsers expecting only grounded/partial/absent won't break; look for the
fallback example and references to status values (the JSON example block and any
sentences enumerating statuses) and apply the chosen approach consistently.
🪄 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: Enterprise
Run ID: 25b3b8ba-c966-41d6-8c49-8b253add4533
📒 Files selected for processing (1)
plugins/docs-tools/skills/docs-workflow-scope-req-audit/SKILL.md
5 tasks
…t and requirements steps Restructure two workflow steps to dispatch one subagent per requirement instead of processing all requirements in a single context window. This prevents context degradation when analyzing tickets with many requirements. Phase 1 — scope-req-audit: - New evidence-classifier agent for per-requirement code evidence classification - Skill restructured: parse → discover repos → fan out (parallel) → merge - Index warming step ensures code-finder index is built once before fanout - Removed orphaned prompts/gap-classification.md (logic embedded in agent) Phase 2 — requirements: - New requirements-discoverer agent for lightweight pass-1 enumeration - requirements-analyst scoped down to single-requirement deep analysis - Skill restructured: discovery → parse skeleton → fan out (parallel) → merge → assemble markdown Output contracts unchanged — evidence-status.json, requirements.md, and step-result.json formats are identical. Downstream steps see no difference. Closes #116 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> rh-pre-commit.version: 2.3.2 rh-pre-commit.check-secrets: ENABLED
A complex requirement with JIRA + multiple PRs + spec + web search can consume 10-12 turns. With retries on failed tool calls, 15 was too tight. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> rh-pre-commit.version: 2.3.2 rh-pre-commit.check-secrets: ENABLED
…at runtime Restore the prompt file as the single source of truth for gap categories and recommended actions. The agent now reads it via the Read tool instead of embedding the logic inline. Bumped maxTurns from 5 to 8 to account for the additional Read turn. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> rh-pre-commit.version: 2.3.2 rh-pre-commit.check-secrets: ENABLED
Per Anthropic docs, the skills: frontmatter injects the full skill
content into the subagent's context at startup. Preloading 5 skills
into agents that run N times in parallel wastes context unnecessarily.
Instead, reference scripts directly via ${CLAUDE_PLUGIN_ROOT} paths
(jira-reader, git-pr-reader, gdoc2md.py are already called this way).
Remove the Skill tool since it's no longer needed.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
rh-pre-commit.version: 2.3.2
rh-pre-commit.check-secrets: ENABLED
Replace bash variable syntax ($FIND_EVIDENCE_SCRIPT, $REPO_PATH) with placeholder syntax (<script>, <repo>) to make it clear these values come from the prompt's CONFIGURATION block, not from shell environment variables. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> rh-pre-commit.version: 2.3.2 rh-pre-commit.check-secrets: ENABLED
Remove the Cursor-specific path affordance for FIND_EVIDENCE_SCRIPT.
Use the standard ${CLAUDE_PLUGIN_ROOT} cross-skill script pattern
consistent with all other agent and skill script references.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
rh-pre-commit.version: 2.3.2
rh-pre-commit.check-secrets: ENABLED
Remove script existence check (find_evidence.py ships with the plugin) and discovery file existence check (just written by our own agent). Keep input validation at the system boundary (--repo, requirements.md) and final output verification (step contract). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> rh-pre-commit.version: 2.3.2 rh-pre-commit.check-secrets: ENABLED
- evidence-classifier: clarify classification precedence (evaluate absent first, then grounded, then partial as catch-all). Covers the edge case where top hit >= grounded but < 2 supporting results. - requirements-discoverer: make Git access conditional on PR/MR URLs being present. JIRA-only discovery proceeds without Git. - requirements-discoverer: align related_tickets step 6 description with the actual JSON output structure (container groups, not flat relation field). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> rh-pre-commit.version: 2.3.2 rh-pre-commit.check-secrets: ENABLED
code-finder is a documented prerequisite — call find_evidence.py
directly at its canonical ${CLAUDE_PLUGIN_ROOT} path instead of
detecting install method or passing script paths through variables.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
rh-pre-commit.version: 2.3.2
rh-pre-commit.check-secrets: ENABLED
…trator invocations Add guardrails so the model respects explicit /docs-orchestrator invocations instead of routing them through docs-workflow-start. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> rh-pre-commit.version: 2.3.2 rh-pre-commit.check-secrets: ENABLED
…--pr, qualify agent names Three changes: 1. resolve_source.py: single repos now clone into code-repo/<repo_name>/ instead of flat into code-repo/. Consistent with multi-repo behavior. 2. resolve_source.py: --repo and --pr accept multiple space-delimited values (nargs="+"). Multiple --repo values each get cloned into their own subdirectory with primary + additional_repos in the result. 3. All workflow step skills: use fully qualified agent names (docs-tools:<agent>) in prose, descriptions, and subagent_type fields to prevent "agent not found" errors from bare name dispatch. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> rh-pre-commit.version: 2.3.2 rh-pre-commit.check-secrets: ENABLED
…script The script was clearing the plugin cache but not updating installed_plugins.json, so Claude Code kept reading the stale version entry. Now patches the registry after checkout. Also defaults --branch to the current git branch and --plugin to docs-tools. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> rh-pre-commit.version: 2.3.2 rh-pre-commit.check-secrets: ENABLED
resolve_source.py now queries the JIRA ticket's git_links and auto-discovered PR URLs when no --repo or --pr is provided. It groups discovered URLs by repo and selects the primary by reference count, failing with an actionable error on a tie. This lets the code-evidence workflow run without explicit --source-code-repo when the ticket has linked PRs or repo URLs. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> rh-pre-commit.version: 2.3.2 rh-pre-commit.check-secrets: ENABLED
aireilly
force-pushed
the
feat/fanout-subagents-per-requirement
branch
from
23c2034 to
1491aaa
Compare
… subgroup regex - requirements-analyst: only require access for systems listed in the requirement's sources array instead of unconditionally requiring JIRA/Git - scope-req-audit: use "status": "absent" (not "error") in fallback entries to preserve the grounded|partial|absent downstream contract - resolve_source: fix GITLAB_REPO_RE to match subgroup paths (org/sub/repo) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> rh-pre-commit.version: 2.3.2 rh-pre-commit.check-secrets: ENABLED
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.
Summary
scope-req-auditandrequirementsworkflow steps to dispatch one subagent per requirement instead of processing all requirements in a single context windowPhase 1: scope-req-audit fanout
agents/evidence-classifier.md— per-requirement code evidence classifier (haiku, maxTurns: 5)skills/docs-workflow-scope-req-audit/SKILL.md— parse → discover repos → warm index → fan out one agent per REQ → merge → writeprompts/gap-classification.md— logic embedded in agent definitionPhase 2: requirements two-pass fanout
agents/requirements-discoverer.md— lightweight pass-1 discovery (JIRA traversal, PR listing, spec identification → JSON skeleton)agents/requirements-analyst.md— scoped down from "all requirements" to "one requirement deep analysis" (returns JSON, maxTurns: 15)skills/docs-workflow-requirements/SKILL.md— discovery → parse skeleton → fan out analyst per REQ → merge → assemble requirements.mdVersion
plugin.jsonbumped to 0.0.57Test plan
scope-req-auditon a ticket with 5+ requirements against a source repo — verifyevidence-status.jsonhas per-requirement classificationsrequirementson a multi-source JIRA ticket — verifyrequirements.mdhas REQ-NNN format with full detailsdiscovery.jsonis written and parseablestep-result.jsonschema unchanged for both stepsdocs-workflow-code-evidencepipeline end-to-end — verify planning step consumes evidence-status.json correctlyCloses #116
🤖 Generated with Claude Code
Summary by CodeRabbit
New Features
Updates
Chores