docs: add scheduled-tasks CONTEXT.md and update routing (#2) #6

cal · 2026-03-04T08:05:04Z

cal commented

2026-03-04 08:05:04 +00:00

Summary

Closes #2 — documents the implemented headless Claude scheduled task system.

scheduled-tasks/CONTEXT.md — new file covering the complete system: runner.sh framework, custom dispatcher pattern, all active tasks, settings.json reference, auth model, monitoring commands, and cost safety guardrails
CLAUDE.md — updated routing table to load the new CONTEXT.md alongside the existing SKILL.md when scheduled-task topics come up; added keywords for issue-poller, issue-worker, pr-reviewer, backlog-triage

What's documented

The system has two execution patterns:

runner.sh template framework — claude-scheduled@<task>.timer → template service → runner.sh reads tasks/<task>/{prompt.md,settings.json,mcp.json} and invokes claude -p
Custom dispatcher scripts — issue-poller.sh, pr-reviewer-dispatcher.sh, daily-report.py with their own service units

Active tasks: backlog-triage (weekdays 09:15), issue-poller (every 30 min), issue-worker (on-demand), pr-reviewer (on timer), daily-report (daily).

Files Changed

scheduled-tasks/CONTEXT.md (new)
CLAUDE.md (routing table update)

Test Results

No test suite — files verified by reading back. CONTEXT.md accurately reflects the live system at ~/.config/claude-scheduled/.

## Summary Closes #2 — documents the implemented headless Claude scheduled task system. - **`scheduled-tasks/CONTEXT.md`** — new file covering the complete system: runner.sh framework, custom dispatcher pattern, all active tasks, `settings.json` reference, auth model, monitoring commands, and cost safety guardrails - **`CLAUDE.md`** — updated routing table to load the new `CONTEXT.md` alongside the existing `SKILL.md` when scheduled-task topics come up; added keywords for `issue-poller`, `issue-worker`, `pr-reviewer`, `backlog-triage` ## What's documented The system has two execution patterns: 1. **runner.sh template framework** — `claude-scheduled@<task>.timer` → template service → `runner.sh` reads `tasks/<task>/{prompt.md,settings.json,mcp.json}` and invokes `claude -p` 2. **Custom dispatcher scripts** — `issue-poller.sh`, `pr-reviewer-dispatcher.sh`, `daily-report.py` with their own service units Active tasks: `backlog-triage` (weekdays 09:15), `issue-poller` (every 30 min), `issue-worker` (on-demand), `pr-reviewer` (on timer), `daily-report` (daily). ## Files Changed - `scheduled-tasks/CONTEXT.md` (new) - `CLAUDE.md` (routing table update) ## Test Results No test suite — files verified by reading back. CONTEXT.md accurately reflects the live system at `~/.config/claude-scheduled/`.

cal added 1 commit 2026-03-04 08:05:05 +00:00

docs: add scheduled-tasks documentation and update CLAUDE.md routing (#2 ) 4ebf834d6a

Adds scheduled-tasks/CONTEXT.md documenting the implemented headless
Claude scheduled task system: runner.sh framework, custom dispatcher
scripts, active tasks, settings reference, auth, monitoring, and
cost safety guardrails.

Updates CLAUDE.md routing table to load CONTEXT.md alongside the
SKILL.md for all scheduled-task-related topics.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

cal referenced this pull request

2026-03-04 08:05:19 +00:00

Scheduled Tasks: headless Claude sessions on a timer (Cowork-style) #2

cal added the

ai-reviewing

label 2026-03-04 08:17:19 +00:00

cal reviewed 2026-03-04 08:18:24 +00:00

cal left a comment

AI Code Review

Files Reviewed

CLAUDE.md (modified)
scheduled-tasks/CONTEXT.md (added)

Findings

Correctness

CLAUDE.md routing update correctly expands the scheduled-task keywords and now points to both scheduled-tasks/CONTEXT.md and the existing SKILL.md — all prior keywords (cowork, timer, etc.) are preserved.
CONTEXT.md accurately describes both execution patterns (runner.sh template framework and custom dispatchers) and matches the PR body's claims.
Active tasks table, file layout, settings reference, and monitoring commands are internally consistent.
The issue-dispatcher.sh entry in the file layout is correctly included (it's a distinct dispatcher from issue-poller.sh), even though the PR body only mentioned three custom scripts — no discrepancy.

Security

No hardcoded credentials. The Discord webhook URL in the settings example uses ... as a placeholder — correct.
Auth model documented clearly: OAuth via Claude Max subscription, no API key.
Cost safety guardrails (budget cap, allowedTools, timeout, kill switch) are accurately documented.

Style & Conventions

Follows the project's CONTEXT.md pattern: clear sections, bash code blocks, tables for reference data.
settings.json table documents both field purpose and defaults — matches the quality of other context files in this repo.
CLAUDE.md line change is minimal and targeted; no unrelated churn.

Suggestions

Consider adding a "Troubleshooting" section or pointer in future (e.g., common failure modes: auth expiry, MCP config errors, budget exhaustion). Not blocking — the monitoring commands cover the basics.

Verdict: APPROVED

Clean, well-structured documentation PR. Content is accurate, conventions are followed, no security concerns. The dual-pointer routing in CLAUDE.md (CONTEXT.md + SKILL.md) is the right pattern — operators get the live system reference while the SKILL.md remains the creation guide.

Note: Gitea blocks self-approval; posting as COMMENT. Recommend merging.

Automated review by Claude PR Reviewer

## AI Code Review ### Files Reviewed - `CLAUDE.md` (modified) - `scheduled-tasks/CONTEXT.md` (added) ### Findings #### Correctness - CLAUDE.md routing update correctly expands the scheduled-task keywords and now points to both `scheduled-tasks/CONTEXT.md` and the existing `SKILL.md` — all prior keywords (`cowork`, `timer`, etc.) are preserved. - CONTEXT.md accurately describes both execution patterns (runner.sh template framework and custom dispatchers) and matches the PR body's claims. - Active tasks table, file layout, settings reference, and monitoring commands are internally consistent. - The `issue-dispatcher.sh` entry in the file layout is correctly included (it's a distinct dispatcher from `issue-poller.sh`), even though the PR body only mentioned three custom scripts — no discrepancy. #### Security - No hardcoded credentials. The Discord webhook URL in the settings example uses `...` as a placeholder — correct. - Auth model documented clearly: OAuth via Claude Max subscription, no API key. - Cost safety guardrails (budget cap, allowedTools, timeout, kill switch) are accurately documented. #### Style & Conventions - Follows the project's CONTEXT.md pattern: clear sections, bash code blocks, tables for reference data. - `settings.json` table documents both field purpose and defaults — matches the quality of other context files in this repo. - CLAUDE.md line change is minimal and targeted; no unrelated churn. #### Suggestions - Consider adding a "Troubleshooting" section or pointer in future (e.g., common failure modes: auth expiry, MCP config errors, budget exhaustion). Not blocking — the monitoring commands cover the basics. ### Verdict: APPROVED Clean, well-structured documentation PR. Content is accurate, conventions are followed, no security concerns. The dual-pointer routing in CLAUDE.md (`CONTEXT.md` + `SKILL.md`) is the right pattern — operators get the live system reference while the SKILL.md remains the creation guide. > Note: Gitea blocks self-approval; posting as COMMENT. Recommend merging. --- *Automated review by Claude PR Reviewer*

cal added

ai-reviewed

and removed

ai-reviewing

labels 2026-03-04 08:18:43 +00:00

cal force-pushed ai/claude-home#2 from 4ebf834d6a to b50808eb8e

2026-03-05 03:55:07 +00:00

Compare

cal merged commit 24acf99836 into main

2026-03-05 03:55:20 +00:00

cal deleted branch ai/claude-home#2

2026-03-05 03:55:21 +00:00