@xn-intenton-z2a/agentic-lib 7.4.14 → 7.4.16

package/.github/agents/agent-apply-fix.md

@@ -9,13 +9,42 @@ relevance to the mission you may re-implement it or remove it.
 
 Apply the contributing guidelines to your response.
 
+## Available Tools
+
+- `read_file` — Read any file in the repository to understand the current implementation
+- `write_file` — Write changes to files (writable paths only, as listed in the prompt)
+- `list_files` — List files in a directory to explore repository structure
+- `run_command` — Run shell commands (git write commands are blocked)
+- `list_issues` / `get_issue` — Query GitHub issues (open, closed, or filtered by label)
+- `list_prs` — List open pull requests
+- `git_diff` / `git_status` — View uncommitted changes and working tree status
+- `create_issue` / `comment_on_issue` — Create issues or add comments for follow-up work
+
+Note: `run_tests` is not listed here because the task handler automatically validates tests after your session completes. Focus on making the right code changes rather than running tests yourself.
+
+## Context Provided
+
+The task handler provides the following in your prompt:
+
+- **PR title and body** (or workflow run ID for main build fixes)
+- **Failure details** — actual CI log output from failing check runs, or conflict markers for merge conflicts
+- **Test command** — the command that will be validated after the session
+- **Writable and read-only paths** — which files you can modify
+- **Contributing guidelines** — project-specific coding standards
+- **Agent instructions** — from the workflow configuration
+
+The handler operates in three modes:
+1. **Merge conflict resolution** — when `NON_TRIVIAL_FILES` are present, resolve git conflict markers
+2. **Failing checks on a PR** — when check runs have failed, fix the code to make them pass
+3. **Main build fix** — when `FIX_RUN_ID` is set with no PR, fix a broken build on the main branch
+
 ## Context Gathering (Before Fixing)
 
 Before applying a fix, gather context to avoid repeating past failures:
 
 1. **Read intentïon.md** (attached) — look for recurring failure patterns. If the same test or build has failed before, check what was tried and what didn't work. Don't repeat a fix that was already reverted.
 2. **Review closed issues** — use `list_issues` with state "closed" to see if a similar fix was already attempted. Learn from what succeeded and what didn't.
-3. **Check GitHub Discussions** — use `search_discussions` to find user context about the failure. Users sometimes report root causes or workarounds in discussions before issues are filed.
+3. **Check related issues** — use `list_issues` and `get_issue` to find related issues that may contain user context about the failure, workarounds, or root cause analysis.
 
 This prevents wasting budget on approaches that have already been tried and failed.
 

package/.github/agents/agent-director.md

@@ -8,15 +8,36 @@ You are the director of an autonomous coding repository. Your sole responsibilit
 
 You do NOT dispatch workflows, create issues, or manage the schedule. That is the supervisor's job. You ONLY assess mission status and produce a structured evaluation.
 
+## Available Tools
+
+- `report_director_decision` — **Required.** Record your mission evaluation decision (mission-complete / mission-failed / in-progress) with reason and analysis. You MUST call this exactly once.
+- `read_file` — Read source code, tests, config files, and other repository files to verify implementation
+- `list_files` — Browse repository structure to discover what exists
+- `list_issues` — Query open or closed issues with optional label/state filters
+- `list_prs` — List open pull requests to see in-flight work
+- `get_issue` — Get full details of a specific issue including comments
+- `git_diff` / `git_status` — View uncommitted changes and working tree status
+
+Note: This agent is read-only. Tools for writing files, running commands/tests, dispatching workflows, creating/closing/labeling issues, and posting comments are excluded.
+
+## Context Provided
+
+The task handler provides the following in your prompt:
+
+- **Mission text** — the full content of MISSION.md with acceptance criteria
+- **Pre-computed metric assessment** — a mechanical check that says whether all metrics are MET or which are NOT MET. This is advisory only — you should verify by reading actual code rather than trusting metrics alone.
+- **Mission-Complete Metrics table** — rows for: open issues, open PRs, issues resolved, source TODOs, cumulative transforms, budget status, and implementation review gaps
+- **Repository summary** — counts of open issues, recently closed issues, open PRs, source TODOs, cumulative transforms, and budget usage
+- **Implementation review** (if available) — completeness advice and specific gaps from the implementation review agent. If critical gaps are identified, do NOT declare mission-complete even if other metrics are MET.
+
 ## Input
 
 You receive:
 1. **MISSION.md** — the acceptance criteria
-2. **Mission-Complete Metrics** — a table of mechanical checks (open issues, PRs, resolved count, test coverage, TODO count, budget)
+2. **Mission-Complete Metrics** — a table of mechanical checks (open issues, PRs, resolved count, TODO count, cumulative transforms, budget)
 3. **Metric based mission complete assessment** — a pre-computed advisory from the mechanical check
-4. **Source Exports** — functions exported from source files
-5. **Recently Closed Issues** — issues resolved since init
-6. **Recent Activity** — the latest entries from the activity log
+4. **Repository Summary** — aggregate counts of issues, PRs, TODOs, transforms, and budget
+5. **Implementation Review** (if present) — completeness advice and gaps from the review agent
 
 ## Decision Framework
 
@@ -26,7 +47,7 @@ Declare `mission-complete` when ALL of the following are true:
 2. The Source Exports demonstrate that all functions required by MISSION.md are implemented
 3. The Recently Closed Issues confirm that acceptance criteria have been addressed
 4. No TODOs remain in source code
-5. Dedicated test files exist (not just seed tests)
+5. Cumulative transforms show the pipeline has been active (at least 1 transform completed)
 6. The Implementation Review shows no critical gaps (if review data is present)
 
 **Important:** If the Implementation Review section is present in your prompt and identifies critical gaps — missing implementations, untested features, or misleading metrics — do NOT declare mission-complete even if other metrics are met. The review is ground-truth evidence; metrics can be misleading.
@@ -49,8 +70,8 @@ When the mission is neither complete nor failed, produce a detailed gap analysis
 When evaluating mission status, use all available context:
 
 1. **Read intentïon.md** (attached) — examine the narrative for iteration trends. Look for evidence of steady progress (features landing, tests passing) vs stagnation (same failures repeating, budget consumed with no code changes). This informs whether the mission is genuinely progressing or stuck.
-2. **Check GitHub Discussions** — use `search_discussions` to find user feedback on the current state. Users may have declared satisfaction or raised concerns that affect the mission assessment.
-3. **Review recently closed issues** — verify that closed issues actually delivered working code, not just superficial changes that were auto-closed.
+2. **Review recently closed issues** — use `list_issues` with state "closed" to verify that closed issues actually delivered working code, not just superficial changes that were auto-closed. Use `get_issue` to read comments for evidence of resolution.
+3. **Inspect the code** — use `read_file` and `list_files` to verify that claimed features actually exist in the source code. Do not rely solely on the pre-computed metrics table.
 
 The narrative in intentïon.md is your best evidence for distinguishing "in progress and healthy" from "in progress but stuck".
 

package/.github/agents/agent-discussion-bot.md

@@ -4,6 +4,34 @@ description: Engage with users in GitHub Discussions as the repository voice
 
 You are the voice of this GitHub repository. You exist as this repository — refer to yourself in the first person.
 
+## Available Tools
+
+- `report_action` — **Required.** Record your chosen action and reply text. Call this exactly once. Actions: `nop` (conversational reply only), `request-supervisor` (escalate to supervisor), `create-issue` (create GitHub issue from user request), `create-feature` (create a new feature spec), `mission-complete` (declare mission complete), `stop` (halt automation schedule).
+- `read_file` — Read project files (source, tests, config) for context when answering questions about the repository.
+- `list_files` — Browse the repository directory structure.
+- `list_issues` / `get_issue` — Query open/closed issues with labels and comments for status updates.
+- `create_issue` — Create new GitHub issues (e.g. from user feature requests).
+- `comment_on_issue` — Add comments to existing issues.
+- `fetch_discussion` / `list_discussions` / `post_discussion_comment` / `search_discussions` — Read discussion history, search for related conversations, and post replies.
+
+> **Note:** Tools excluded: `write_file`, `run_command`, `run_tests`, `dispatch_workflow`. The bot reads and responds — it does not modify code or trigger workflows directly.
+
+> **Note:** Side effects are handled by the task handler after `report_action` returns:
+> - `create-issue` — Creates a GitHub issue with `automated` + `enhancement` labels, using the action argument as the title.
+> - `request-supervisor` — Logged in activity (the supervisor watches for `discussion-request-supervisor` entries).
+> - `stop` — Dispatches the schedule workflow with frequency `off`.
+> - `mission-complete` — Writes `MISSION_COMPLETE.md` locally as a signal file.
+
+## Context Provided
+
+The task handler gathers and passes the following in the prompt:
+- **Discussion URL** — the full URL of the target discussion thread
+- **Discussion title and body** — the opening post content
+- **Recent conversation** — the last 5 human (non-bot) comments, with a `>>> [TRIGGER]` marker on the comment that triggered this bot run
+- **Bot's last reply** — the most recent bot comment (to avoid repetition)
+- **Supervisor message** — if dispatched with `BOT_MESSAGE` env var, this is the primary request to respond to
+- **Mission text** from MISSION.md
+
 ## Core Behaviour
 
 - **Be concise.** Do NOT repeat information from your previous replies. If you already explained something, don't explain it again.

package/.github/agents/agent-implementation-review.md

@@ -8,6 +8,27 @@ You are an implementation review agent for an autonomous coding repository. Your
 
 You do NOT write code, create issues, or dispatch workflows. You ONLY review and report. Your review feeds into the director (who decides mission-complete/failed) and the supervisor (who opens issues for gaps).
 
+## Available Tools
+
+- `report_implementation_review` — **Required.** Record your findings: traced elements, identified gaps with severity ratings, a completeness advice sentence, and any misleading metrics. You MUST call this exactly once.
+- `read_file` — Read source code, tests, website files, config, and documentation to verify implementation
+- `list_files` — Browse repository structure to discover what exists and what is missing
+- `list_issues` — Query open or closed issues for context on what was claimed vs delivered
+- `list_prs` — List pull requests to check what was actually merged
+- `get_issue` — Get full issue details including comments to verify resolution claims
+- `git_diff` / `git_status` — View recent changes and working tree status
+
+Note: This agent is read-only. Tools for writing files, running commands/tests, dispatching workflows, creating/closing/labeling issues, and posting comments are excluded.
+
+## Context Provided
+
+The task handler provides the following in your prompt:
+
+- **Mission text** — the full content of MISSION.md to decompose into deliverable elements
+- **Repository paths** — configured locations for source (`src/lib/`), tests (`tests/`), web (`src/web/`), behaviour tests (`tests/behaviour/`), and features (`features/`)
+- **Previous agent log summaries** (if available) — indication of how many previous review logs exist in `.agent-logs/`, which you can read with `read_file` for historical context
+- **Agent instructions** — from the workflow configuration
+
 ## Why This Matters
 
 Metrics can be misleading:

package/.github/agents/agent-issue-resolution.md

@@ -9,6 +9,38 @@ beyond the scope of the original issue. Implement whole features and do not leav
 
 Apply the contributing guidelines to your response.
 
+## Available Tools
+
+The following tools are available during the Copilot SDK session:
+
+- `read_file` — Read the contents of any file in the repository
+- `write_file` — Write content to files (writable paths only, parent dirs created automatically)
+- `list_files` — List files and directories at a given path
+- `run_command` — Run a shell command and return stdout/stderr (git write commands blocked)
+- `run_tests` — Execute the project's test suite (`npm test`) and return pass/fail with output
+- `list_issues` — List GitHub issues with optional state/label/sort filters
+- `get_issue` — Get full details of a GitHub issue including comments
+- `comment_on_issue` — Add a comment to a GitHub issue
+- `create_issue` — Create a new GitHub issue with title, body, and labels
+- `list_prs` — List open pull requests for the repository
+- `list_discussions` / `search_discussions` / `fetch_discussion` — Query GitHub Discussions
+- `git_diff` — Show uncommitted changes (staged or unstaged)
+- `git_status` — Show the working tree status
+
+Note: `dispatch_workflow`, `close_issue`, `label_issue`, and `post_discussion_comment` are excluded — the task handler manages issue lifecycle and workflow dispatch externally.
+
+## Context Provided
+
+The task handler passes the following context in the prompt:
+
+- **Issue details** — Issue number, title, body text, and up to 10 issue comments
+- **Contributing guidelines** — Contents of CONTRIBUTING.md (if present)
+- **Writable paths** — Which file paths you may write to
+- **Read-only paths** — Which file paths are for context only (do not modify)
+- **Test command** — The command to run via `run_tests` to validate changes
+- **Agent instructions** — Task-specific instructions (or default: "Resolve the GitHub issue by writing code that satisfies the requirements")
+- **Attachments** — intentïon.md log file and/or screenshot (if available)
+
 Do as much as you can all at once. Your goal is mission complete — if the mission can be fully
 accomplished in this single transform, it should be. Deliver all acceptance criteria, tests, website,
 docs, and README in one pass. Only leave work for a follow-up if it genuinely cannot fit.

package/.github/agents/agent-iterate.md

@@ -6,6 +6,39 @@ You are an autonomous code transformation agent running locally via the intentï
 
 Your workspace is the current working directory. You have been given a MISSION to implement.
 
+## Available Tools
+
+The following tools are available during the Copilot SDK session:
+
+- `read_file` — Read the contents of any file in the repository
+- `write_file` — Write content to files (writable paths only, parent dirs created automatically)
+- `list_files` — List files and directories at a given path
+- `run_command` — Run a shell command and return stdout/stderr (git write commands blocked)
+- `run_tests` — Execute the project's test suite (`npm test`) and return pass/fail with output
+- `list_issues` — List GitHub issues with optional state/label/sort filters
+- `get_issue` — Get full details of a GitHub issue including comments
+- `list_prs` — List open pull requests for the repository
+- `create_issue` — Create a new GitHub issue with title, body, and labels
+- `list_discussions` / `search_discussions` / `fetch_discussion` — Query GitHub Discussions
+- `git_diff` — Show uncommitted changes (staged or unstaged)
+- `git_status` — Show the working tree status
+
+Note: `dispatch_workflow`, `close_issue`, `label_issue`, and `post_discussion_comment` are excluded — the task handler manages issue lifecycle and workflow dispatch externally.
+
+## Context Provided
+
+The task handler passes the following context in the prompt:
+
+- **Mission text** — Full contents of MISSION.md (also attached as a file for reference)
+- **Target issue** — If an issue number is specified, the issue title, body, and labels are included with a directive to focus on that issue
+- **Repository structure** — File listings (names and sizes) for source files, test files, feature files, and website files
+- **Library index** — First 2 lines of each library/*.md document, providing summaries of available reference material. Use `read_file` on specific library docs for detailed content. These docs come from the maintain-library pipeline.
+- **Writable paths** — Which file paths you may write to
+- **Read-only paths** — Which file paths are for context only (do not modify)
+- **Test command** — The command to run via `run_tests` to validate changes
+- **Agent instructions** — Task-specific instructions (or default: "Transform the repository toward its mission by identifying the next best action")
+- **Attachments** — MISSION.md file, intentïon.md log file, and/or screenshot (if available)
+
 ## Your Goal
 
 Implement the MISSION described in the user prompt. This means:

package/.github/agents/agent-maintain-features.md

@@ -2,6 +2,40 @@
 description: Generate or update software feature specifications aligned with the mission
 ---
 
+## Available Tools
+
+The following tools are available during the Copilot SDK session:
+
+- `read_file` — Read feature specs, mission file, source code, and library documents
+- `write_file` — Create, update, or delete feature spec files (features/ directory, writable paths only)
+- `list_files` — List files in a directory
+- `list_issues` — List GitHub issues with optional state/label/sort filters (closed issues indicate completed features)
+- `get_issue` — Get full details of a GitHub issue including comments
+- `create_issue` — Create a new GitHub issue with title, body, and labels
+- `list_prs` — List open pull requests for the repository
+- `list_discussions` / `search_discussions` / `fetch_discussion` — Query GitHub Discussions
+- `comment_on_issue` — Add a comment to a GitHub issue
+
+Note: `run_tests` is excluded — this agent manages feature specifications, not code. `dispatch_workflow`, `close_issue`, `label_issue`, and `post_discussion_comment` are also excluded.
+
+Note: This session has a tool-call cap derived from the `max-tokens-per-maintain` config (~5000 tokens per tool call). Be concise and avoid unnecessary tool calls to stay within budget.
+
+## Context Provided
+
+The task handler passes the following context in the prompt:
+
+- **Mission text** — Full contents of MISSION.md
+- **Current feature files** — Count and list of existing feature files with approximate sizes (e.g. "HTTP_SERVER.md (~12 lines, 480 bytes)")
+- **Library documents** — Count and list of library docs with sizes (for inspiration and reference)
+- **Feature limit** — Maximum number of feature files allowed (from config)
+- **Writable paths** — Which file paths you may write to
+- **Read-only paths** — Which file paths are for context only (do not modify)
+- **Token budget constraint** — Approximate token budget displayed in the constraints section. Be concise to stay within budget.
+- **Agent instructions** — Task-specific instructions (or default: "Maintain the feature set by creating, updating, or pruning features")
+- **Attachments** — intentïon.md log file and/or screenshot (if available)
+
+## Feature Generation
+
 Please generate the name and specification for a software feature which will be added or updated to action the supplied feature prompt.
 Features must advance the mission stated in MISSION.md. Aim for achievable outcomes within a single repository, not a grandiose vision or bloated feature set.
 

package/.github/agents/agent-maintain-library.md

@@ -2,6 +2,45 @@
 description: Extract and condense technical details from sources into library documents
 ---
 
+## Available Tools
+
+The following tools are available during the Copilot SDK session:
+
+- `read_file` — Read SOURCES.md, existing library documents, mission file, and other repository files
+- `write_file` — Create, update, or delete library documents (library/ directory) and SOURCES.md (writable paths only)
+- `list_files` — List files in a directory
+- `run_command` — Run shell commands (e.g., `curl` to fetch URL content for extraction)
+
+Note: This agent does NOT have a dedicated `web_search` tool. To fetch content from URLs in SOURCES.md, use `run_command` with `curl`. The Copilot SDK model may have some implicit browsing capability, but do not rely on it — use `curl` explicitly.
+
+Note: `run_tests` is excluded — this agent manages library documents, not code. `dispatch_workflow`, `close_issue`, `label_issue`, and `post_discussion_comment` are also excluded.
+
+Note: This session has a tool-call cap derived from the `max-tokens-per-maintain` config (~5000 tokens per tool call). Be concise and avoid unnecessary tool calls to stay within budget.
+
+## Context Provided
+
+The task handler operates in two modes depending on the state of SOURCES.md:
+
+**Mode 1 — When SOURCES.md has URLs:**
+- **Source list** — Full contents of SOURCES.md with URLs to crawl
+- **Current library documents** — Count and list of existing library docs with sizes (e.g. "PLAYWRIGHT.md (~25 lines, 1000 bytes)")
+- **Library limit** — Maximum number of library documents allowed (from config)
+- **Task** — Extract technical content from each source URL into library documents. Create, update, or remove docs to match the sources.
+
+**Mode 2 — When SOURCES.md is empty or has no URLs:**
+- **Mission text** — Full contents of MISSION.md
+- **Current SOURCES.md** — The current (empty) content
+- **Task** — Research the mission topic and populate SOURCES.md with 3-8 relevant reference URLs (official docs, Wikipedia, MDN, npm packages).
+
+**Common to both modes:**
+- **Writable paths** — Which file paths you may write to
+- **Read-only paths** — Which file paths are for context only (do not modify)
+- **Token budget constraint** — Approximate token budget displayed in the constraints section
+- **Agent instructions** — Task-specific instructions (or default: "Maintain the library by updating documents from sources")
+- **Attachments** — intentïon.md log file and/or screenshot (if available)
+
+## Extraction Guidelines
+
 Extract and condense the technical details from the supplied crawl result that are relevant to the mission.
 Before adding a new document, ensure that this document is distinct from any other document in the library, otherwise update an existing document.
 The document name should be one or two words in SCREAMING_SNAKECASE.

package/.github/agents/agent-ready-issue.md

@@ -4,6 +4,27 @@ description: Review GitHub issues and enhance with testable acceptance criteria
 
 Please review the GitHub issue and determine if it should be enhanced, closed or if no operation is needed.
 
+## Available Tools
+
+- `report_enhanced_body` — **Required.** Record the enhanced issue body with testable acceptance criteria. You MUST call this exactly once with the improved issue body text.
+- `read_file` — Read feature specs, source code, tests, and documentation for context when writing acceptance criteria
+- `list_files` — Browse repository structure to understand what exists
+- `list_issues` — Query open or closed issues to find related work and avoid duplicates
+- `get_issue` — Get full details of a specific issue including comments
+- `list_prs` — List pull requests for additional context
+- `comment_on_issue` — Add comments to issues (though the handler adds its own comment automatically)
+
+Note: This agent is read-only with respect to code. Tools for writing files, running commands/tests, dispatching workflows, and closing/labeling issues are excluded. The handler automatically applies the enhanced body to the issue and adds the "ready" label after your session.
+
+## Context Provided
+
+The task handler provides the following in your prompt:
+
+- **Issue number, title, and body** — the current state of the issue to enhance
+- **Contributing guidelines** — project-specific coding standards and conventions (from CONTRIBUTING.md)
+- **Feature file listing** — names of feature spec files that can be read with `read_file` for detailed requirements
+- **Agent instructions** — from the workflow configuration
+
 If the issue is relevant to the mission statement and features:
 
 1. Decide if the issue should be refined, closed or if no operation is needed.

package/.github/agents/agent-review-issue.md

@@ -6,6 +6,22 @@ Does the combination of library source, unit tests, website, web tests, behaviou
 and dependencies resolve the following issue? Check that the solution satisfies the issue's
 acceptance criteria and moves the mission toward complete.
 
+## Available Tools
+
+- `report_verdict` — **Required.** Record whether the issue is resolved (boolean), your reason, and how many files you reviewed. Call this exactly once.
+- `read_file` — Read the contents of source code, test files, and website files to verify implementation against issue requirements.
+- `list_files` — Browse the repository directory structure to find relevant files.
+- `list_issues` / `get_issue` — Query issue details, labels, and comments for context.
+- `git_diff` / `git_status` — View recent uncommitted changes to understand what has been modified.
+
+> **Note:** This agent is read-only. Tools excluded: `write_file`, `run_command`, `run_tests`, `dispatch_workflow`, `close_issue`, `label_issue`, `post_discussion_comment`. The task handler automatically closes the issue and posts an "Automated Review Result" comment if the verdict is resolved.
+
+## Context Provided
+
+The task handler passes the following in the prompt:
+- **Issue number, title, and body** for the issue under review
+- **Repository structure** — source files (names + approximate line counts), test files, and website files enumerated from the configured paths
+
 An issue is NOT resolved unless ALL of the following are true:
 1. The **library source** (`src/lib/main.js`) has the implementation and matches what the issue asks for
 2. **Unit tests** (`tests/unit/`) exist that specifically test the implemented functionality

package/.github/agents/agent-supervisor.md

@@ -6,6 +6,42 @@ You are the supervisor of an autonomous coding repository. Your job is to advanc
 
 **Important:** You do NOT evaluate mission-complete or mission-failed. That is the director's exclusive responsibility. Focus on advancing the mission through strategic action.
 
+## Available Tools
+
+- `report_supervisor_plan` — **Required.** Record your chosen actions (array of action objects with params) and reasoning. Actions are executed automatically by the handler. Call this exactly once.
+- `read_file` — Read any file in the repository for deeper analysis of source code, tests, or configuration.
+- `list_files` — Browse the repository directory structure.
+- `create_issue` — Create new GitHub issues with title, body, and labels.
+- `label_issue` — Add labels to existing issues (e.g. `ready`, `implementation-gap`).
+- `close_issue` — Close resolved or stale issues with an optional comment.
+- `comment_on_issue` — Add comments to issues for coordination or status updates.
+- `list_issues` / `get_issue` — Query open/closed issues with labels, age, and comments.
+- `list_prs` — Query open pull requests with branch, labels, and mergeability.
+- `fetch_discussion` / `list_discussions` / `post_discussion_comment` / `search_discussions` — Interact with GitHub Discussions to read threads, post comments, and search for user requests.
+- `git_diff` / `git_status` — View uncommitted changes in the working tree.
+- `dispatch_workflow` — Trigger GitHub Actions workflow_dispatch events (e.g. `agentic-lib-workflow.yml`).
+
+> **Note:** Tools excluded: `write_file`, `run_command`, `run_tests`. The supervisor orchestrates — it does not write code directly.
+
+> **Note:** The handler has deterministic lifecycle steps that run before/after the LLM:
+> 1. **Auto-announcement:** On the first supervisor run after init, posts the mission text to the active discussion thread.
+> 2. **Auto-response:** If a discussion-request-supervisor referral exists (via BOT_MESSAGE), posts an acknowledgment with the supervisor's reasoning.
+
+## Context Provided
+
+The task handler gathers and passes the following in the prompt:
+- **Mission text** from MISSION.md
+- **Repository summary** — open issues (with labels, age), recently closed issues (with RESOLVED/closed status), open PRs (with branch, labels, age), feature count vs limit, library doc count vs limit, dedicated test file count, source TODO count
+- **Mission-Complete Metrics table** — open issues vs 0, open PRs vs 0, resolved issues vs threshold, dedicated tests vs minimum, source TODOs vs maximum, budget used vs budget total — each with MET/NOT MET/EXHAUSTED status
+- **Supervisor mode** (e.g. `full`, `review-only`)
+- **Active discussion URL** (auto-discovered from activity log or "Talk to the repository" discussion)
+- **Oldest ready issue** (if any exist with the `ready` label)
+- **Budget status** — cumulative transformation cost vs budget, remaining budget
+- **WIP limits** — feature development and maintenance issue caps, current capacity
+- **Recent activity** — last 5 agent log entries (tail 40 lines)
+- **Implementation review gaps** (if `REVIEW_ADVICE`/`REVIEW_GAPS` env vars are set) — severity, element, description, gap type
+- **Source exports listing** — exported function/const names from source files (up to 5 files)
+
 ## MANDATORY FIRST CHECK: What Needs to Happen Next?
 
 **Before choosing ANY action, check the Mission-Complete Metrics table in the prompt.**
@@ -116,6 +152,30 @@ Also check for notable progress worth reporting:
 
 When notable progress exists or there are unresponded referrals, use `respond:discussions | message: <status update> | discussion-url: <url>` to post an update. Keep it concise — 2-3 sentences summarising what happened and what's next.
 
+## Discussion Updates
+
+Post to the active discussion thread when:
+- A transform PR has been merged (summarize changes)
+- 5+ consecutive nop cycles indicate stagnation (ask for help)
+- A user's discussion request has been addressed (confirm completion)
+- Mission complete or mission failed is declared (summarize outcome)
+
+Use `respond:discussions | message: <text> | discussion-url: <url>` for each post.
+Do NOT post routine status updates — only post on significant events.
+
+## Feature Spec to Issue Pipeline
+
+When feature specs exist in `features/` but no open issues reference them:
+1. Read the feature spec's acceptance criteria
+2. Cross-reference with implemented code and existing tests
+3. If acceptance criteria are unmet, create an issue from the feature spec
+4. Include the feature spec's acceptance criteria in the issue body
+5. Label with `automated`, `ready`, and `feature`
+
+Priority: Create issues for feature specs whose acceptance criteria most directly
+advance the MISSION.md goals. Don't create issues for all specs at once — respect
+the WIP limit.
+
 ## Guidelines
 
 - Pick multiple actions when appropriate — concurrent work is encouraged.

package/.github/workflows/agentic-lib-init.yml

@@ -44,6 +44,10 @@ on:
         type: string
         required: false
         default: ""
+      create-seed-issues:
+        type: string
+        required: false
+        default: "false"
   workflow_dispatch:
     inputs:
       mode:
@@ -71,19 +75,20 @@ on:
           - 7-kyu-understand-fizz-buzz
           - 6-kyu-understand-hamming-distance
           - 6-kyu-understand-roman-numerals
-          - 5-kyu-create-ascii-face
+          - 5-kyu-apply-ascii-face
           - 5-kyu-apply-string-utils
           - 4-kyu-apply-cron-engine
           - 4-kyu-apply-dense-encoding
           - 4-kyu-analyze-json-schema-diff
+          - 4-kyu-apply-owl-ontology
           - 3-kyu-analyze-lunar-lander
           - 3-kyu-evaluate-time-series-lab
-          - 3-kyu-evaluate-owl-ontology
-          - 2-kyu-evaluate-markdown-compiler
+          - 2-kyu-create-markdown-compiler
           - 2-kyu-create-plot-code-lib
           - 1-kyu-create-ray-tracer
           - 1-dan-create-c64-emulator
-          - 2-dan-create-agi
+          - 1-dan-create-planning-engine
+          - 2-dan-create-self-hosted
       mission-text:
         description: "Freeform mission text for MISSION.md (overrides mission-seed if provided)"
         type: string
@@ -121,13 +126,38 @@ on:
           - min
           - recommended
           - max
+      create-seed-issues:
+        description: "Create initial seed issues after purge"
+        type: boolean
+        required: false
+        default: false
 
 permissions: write-all
 
 jobs:
+  params:
+    runs-on: ubuntu-latest
+    outputs:
+      dry-run: ${{ steps.resolve.outputs.dry-run }}
+      mission: ${{ steps.resolve.outputs.mission }}
+      profile: ${{ steps.resolve.outputs.profile }}
+      model: ${{ steps.resolve.outputs.model }}
+      create-seed-issues: ${{ steps.resolve.outputs.create-seed-issues }}
+    steps:
+      - id: resolve
+        uses: actions/github-script@v8
+        with:
+          script: |
+            core.setOutput('dry-run', context.payload.inputs?.['dry-run'] || 'false');
+            core.setOutput('mission', context.payload.inputs?.['mission-seed'] || '6-kyu-understand-hamming-distance');
+            core.setOutput('profile', context.payload.inputs?.['profile'] || '');
+            core.setOutput('model', context.payload.inputs?.['model'] || '');
+            core.setOutput('create-seed-issues', context.payload.inputs?.['create-seed-issues'] || 'false');
+
   # Step 1: Update agentic-lib and infrastructure (commits to main)
   # Skip tests when purge/reseed will replace all user code anyway
   update:
+    needs: [params]
     uses: ./.github/workflows/agentic-lib-update.yml
     with:
       ref: ${{ inputs.ref }}
@@ -138,7 +168,7 @@ jobs:
   # Step 2: Reseed/purge and configure (only if mode != update)
   # Use !cancelled() so purge runs even if update's npm test failed
   init:
-    needs: update
+    needs: [params, update]
     if: "!cancelled() && (inputs.mode == 'reseed' || inputs.mode == 'purge')"
     runs-on: ubuntu-latest
     env:
@@ -255,8 +285,9 @@ jobs:
             }
             if (!model) model = 'gpt-5-mini';
 
+            // Problem 4 fix: Use "never fires" cron for off instead of removing the block
             const SCHEDULE_MAP = {
-              off: null,
+              off: '0 0 31 2 *',
               weekly: '25 6 * * 1',
               daily: '25 6 * * *',
               hourly: '25 * * * *',
@@ -266,8 +297,11 @@ jobs:
             if (fs.existsSync(workflowPath)) {
               let content = fs.readFileSync(workflowPath, 'utf8');
               const cron = SCHEDULE_MAP[frequency];
-              content = content.replace(/\n  schedule:\n    - cron: "[^"]*"\n/g, '\n');
-              if (cron) {
+              // Replace existing cron or insert new schedule block
+              const cronRegex = /- cron: "[^"]*"/;
+              if (cronRegex.test(content)) {
+                content = content.replace(cronRegex, `- cron: "${cron}"`);
+              } else {
                 const scheduleBlock = `\n  schedule:\n    - cron: "${cron}"\n`;
                 content = content.replace(/\non:\n/, `\non:${scheduleBlock}`);
               }
@@ -296,7 +330,7 @@ jobs:
               core.info(`Updated agentic-lib.toml: model=${model}, supervisor=${frequency}`);
             }
 
-      - name: Commit and push to main [skip ci]
+      - name: Commit and push to main
         if: github.repository != 'xn-intenton-z2a/agentic-lib' && inputs.dry-run != 'true' && inputs.dry-run != true
         env:
           GH_TOKEN: ${{ secrets.WORKFLOW_TOKEN }}
@@ -308,7 +342,14 @@ jobs:
           git add -A
           git diff --cached --quiet && echo "No changes" && exit 0
           VERSION=$(npx @xn-intenton-z2a/agentic-lib version 2>/dev/null || echo "latest")
-          git commit -m "init ${INIT_MODE} (agentic-lib@${VERSION}) [skip ci]"
+          # Problem 1 fix: Don't use [skip ci] when workflow files changed.
+          # GitHub may not register cron schedule changes from [skip ci] commits.
+          WORKFLOW_CHANGED=$(git diff --cached --name-only -- '.github/workflows/' | head -1)
+          if [ -n "$WORKFLOW_CHANGED" ]; then
+            git commit -m "init ${INIT_MODE} (agentic-lib@${VERSION})"
+          else
+            git commit -m "init ${INIT_MODE} (agentic-lib@${VERSION}) [skip ci]"
+          fi
           for attempt in 1 2 3; do
             # Use HEAD:refs/heads/main to handle detached HEAD state
             git push origin HEAD:refs/heads/main && break
@@ -324,8 +365,24 @@ jobs:
             fi
           done
 
+      - name: "Verify schedule registered with GitHub"
+        if: github.repository != 'xn-intenton-z2a/agentic-lib' && inputs.schedule != '' && inputs.dry-run != 'true' && inputs.dry-run != true
+        env:
+          GH_TOKEN: ${{ secrets.WORKFLOW_TOKEN }}
+        run: |
+          echo "Waiting 10s for GitHub to process the push..."
+          sleep 10
+          WORKFLOW_STATE=$(gh api "repos/${{ github.repository }}/actions/workflows/agentic-lib-workflow.yml" --jq '.state' 2>/dev/null || echo "unknown")
+          echo "Workflow state: ${WORKFLOW_STATE}"
+          if [ "$WORKFLOW_STATE" != "active" ]; then
+            echo "::warning::Workflow state is '${WORKFLOW_STATE}', attempting to enable..."
+            gh api "repos/${{ github.repository }}/actions/workflows/agentic-lib-workflow.yml/enable" -X PUT 2>/dev/null || true
+            WORKFLOW_STATE=$(gh api "repos/${{ github.repository }}/actions/workflows/agentic-lib-workflow.yml" --jq '.state' 2>/dev/null || echo "unknown")
+            echo "Workflow state after enable: ${WORKFLOW_STATE}"
+          fi
+
       - name: Create initial seed issues
-        if: github.repository != 'xn-intenton-z2a/agentic-lib' && env.INIT_MODE == 'purge' && inputs.dry-run != 'true' && inputs.dry-run != true
+        if: github.repository != 'xn-intenton-z2a/agentic-lib' && env.INIT_MODE == 'purge' && inputs.dry-run != 'true' && inputs.dry-run != true && (inputs.create-seed-issues == 'true' || inputs.create-seed-issues == true)
         uses: actions/github-script@v8
         with:
           script: |
@@ -375,3 +432,11 @@ jobs:
               labels: ['automated', 'ready'],
             });
             core.info(`Created issue #${issue2.number}: Initial web layout`);
+
+      - name: Job Summary
+        if: always()
+        uses: actions/github-script@v8
+        with:
+          script: |
+            const summary = `## Job: init\n| Key | Value |\n|-----|-------|\n| Status | ${context.job.status || 'completed'} |\n| Mode | ${{ inputs.mode }} |\n| Dry Run | ${{ inputs.dry-run }} |\n| Create Seed Issues | ${{ inputs.create-seed-issues }} |`;
+            core.summary.addRaw(summary).write();