Cross Channel Daily Review

Create a simple, repeatable daily review workflow across whatever conversation surfaces are available. Use when a user wants one raw note per source, one mer...

Security Scan

The skill's code, runtime instructions, and file I/O are coherent with its stated purpose: it discovers channels from local session transcripts, writes per-channel raw notes, synthesizes a cross-channel report, updates a local index, and performs verification — it does not request credentials or reach out to external endpoints.

Purpose & Capability

The name/description (cross-channel daily review) matches the included scripts and SKILL.md: discovery, normalization, raw file writing, synthesis, index updates, boss summary generation, verification, and retention. There are no unexpected environment variables, cloud credentials, or unrelated binaries requested.

Instruction Scope

SKILL.md instructs the agent to discover channels from sessions/transcripts, collect evidence, write files under memory/daily-review, synthesize reports, update index.json, and verify outputs. The bundled scripts implement those steps and only reference local paths and inputs (sessions_dir, sessions_list_json, workspace). The skill reads local transcript/session data (expected for this purpose); it does not instruct network exfiltration or access unrelated system paths.

Install Mechanism

No install spec is provided (instruction-only with included scripts). All code is local Python scripts; nothing is downloaded or installed from external URLs or package registries.

Credentials

The skill declares no required environment variables, no primary credential, and no config paths. The scripts operate on local files and arguments passed at runtime (workspace, sessions_dir, optional sessions_list_json). There are no unexpected SECRET/TOKEN/KEY requirements.

Persistence & Privilege

The skill does not request an 'always' privilege; user-invocable and autonomous invocation are default platform behaviors. The skill writes outputs into a workspace (memory/daily-review) which is consistent with its purpose; it does not modify other skills or system-wide agent settings.

Assessment

This skill appears coherent and self-contained, but before installing consider: 1) It reads local session transcripts and session lists (sessions_dir / sessions_list_json) and will write review files and index entries under the provided workspace — ensure those paths contain only data you want processed. 2) Review retention/archive behavior and ensure you understand where files are written and how long they are kept; the skill intentionally leaves deletion disabled by default. 3) Delivery (boss-primary / broadcast) only occurs if you configure a delivery destination; check that no external delivery hooks are enabled before running in production. 4) Test first in a safe workspace using generate-only or sample transcripts to verify output formats and that sensitive PII is handled appropriately. 5) Because the skill runs local Python scripts, run them in a controlled environment (virtualenv/isolated workspace) to limit accidental exposure of other files.

Overview

Use this skill to run a reusable, channel-agnostic daily review workflow:

1. Discover or accept a set of channels.
2. Create one raw review file per channel.
3. Generate one synthesized cross-channel review.
4. Update a machine-readable index.
5. Generate a boss-mode summary for one primary delivery channel.
6. Verify outputs before claiming success.

Default design:

• Primary destination: use the configured preferred destination if present
• Delivery mode: boss-primary
• Review window: last 24 hours
• Missing sources: record explicitly; never silently skip them.

When to use this skill

Use this skill when the user asks for any of the following:

• “做全渠道每日复盘 / daily cross-channel review”
• “把不同渠道分别复盘，再统一汇总”
• “老板模式摘要 / boss summary / management summary”
• “把 raw、synthesized、index 都生成出来”
• “让这个流程可复用、可发布、可适配不同渠道”
• “每天固定时间自动复盘并推送”

Workflow

Step 1 — Resolve channels

First decide the channel set:

• Prefer auto-discovery from available sessions, delivery contexts, or user-provided channel list.
• If the user explicitly names channels, trust that list.
• Do not hardcode a small fixed list of apps as the only valid sources.
• Normalize source names to lowercase slugs that match the environment.

Read references/channel-adapter-spec.md before implementing or extending channel handling.

Step 2 — Collect and normalize channel evidence

For each candidate channel, gather:

• time window
• source references (session path, session key, channel id, or equivalent)
• conversational scope inside the channel (DM / group / thread / room / topic)
• participant shape (single bot, multi-bot, workflow group chat, etc.) when knowable
• a concise raw excerpt or evidence summary
• collection status: active | configured | missing | collection_failed

Normalize each channel into the common object shape described in references/channel-adapter-spec.md.

Important:

• do not assume one channel has only one bot
• do not assume one channel maps to one conversation
• support multiple bots and multiple sessions inside the same channel scope

Step 3 — Write raw review files

Write one file per channel under:

memory/daily-review/raw/YYYY/MM/<channel>_YYYY-MM-DD.md

Use scripts/write_raw_reviews.py with normalized JSON when possible.

Rules:

• Always write one file per resolved channel.
• If a channel has no confirmed data, still write a file with 状态：未检出 or failure reason.
• Include the checked sources. Do not silently omit missing channels.

Template: assets/raw-template.md

Step 4 — Generate the synthesized review

Create exactly one synthesized report for the date:

memory/daily-review/synthesized/YYYY/MM/YYYY-MM-DD.md

The report must cover these 5 dimensions:

1. ❌ 做错了什么
2. ⚠️ 没做好的地方
3. ✅ 做得好的地方
4. 📝 遗漏事项
5. 💡 改进建议

Use assets/synthesized-template.md and references/review-dimensions.md.

Step 5 — Update the index

Update or append the record in:

memory/daily-review/index.json

Prefer scripts/update_index.py.

Index records should capture at least:

• date
• active channels
• missing channels
• synthesized file path
• boss summary path (if built)
• verification status

Step 6 — Build boss-mode summary

Build a shorter management summary from the synthesized report.

Default behavior:

• delivery_mode = boss-primary
• boss_channel = primary

If the user configured a preferred destination, use it. If that destination is unavailable, fallback to the first verified available destination and state that clearly.

Use scripts/build_boss_summary.py and assets/boss-summary-template.md.

Step 7 — Verify before reporting success

Run verification with scripts/verify_outputs.py before claiming completion.

Minimum checks:

• raw files exist for all resolved channels
• synthesized file exists and is non-empty
• index updated successfully
• boss summary exists if requested
• delivery target resolved successfully

Never report “done” without verification output.

Delivery modes

Read references/delivery-modes.md when choosing output behavior.

Supported modes:

• boss-primary — send only to one primary channel; default
• broadcast — send to multiple channels
• generate-only — write outputs but do not push externally

Formatting rules

Read references/formatting-standard.md when generating user-visible summaries.

Hard rules:

• Keep sections clearly separated.
• One major numbered point per paragraph block.
• Do not collapse everything into one dense wall of text.
• Put the final action checklist in a standalone text code block when needed.

Failure handling

Read references/failure-handling.md when anything is missing or degraded.

Important:

• Distinguish missing from collection_failed.
• Distinguish “no recent data” from “channel not configured”.
• If delivery falls back from a preferred destination to another verified destination, state it explicitly.

Resources

scripts/

• discover_channels.py — scan real OpenClaw session transcripts and produce channel candidates
• score_discovery_confidence.py — attach confidence scores to channel/scope inference results
• normalize_channel_data.py — normalize channel metadata to one schema
• write_raw_reviews.py — write per-channel raw review markdown files
• update_index.py — create/update index.json (daily / weekly / monthly)
• synthesize_review.py — generate one synthesized cross-channel review from normalized scope-aware data
• render_periodic_summary.py — generate weekly or monthly rollup summaries
• plan_retention.py — produce retention/archive candidates from index policy
• verify_retention_readiness.py — check whether a month is ready for archive/cleanup
• archive_daily_layer.py — archive eligible daily raw / synthesized / boss outputs into archive tree
• mark_archived_records.py — mark archived daily records inside index metadata
• render_cron_plan.py — render the recommended cron automation chain for this skill
• run_retention_cycle.py — run planner + readiness + archive + mark as one lifecycle step
• build_boss_summary.py — render management summary markdown from synthesized metadata
• resolve_delivery_target.py — resolve requested destination with safe fallback
• verify_outputs.py — verify expected outputs exist and are non-empty

references/

• architecture.md — system model and data flow
• channel-adapter-spec.md — normalized channel schema and adapter contract
• review-dimensions.md — criteria for the 5 review dimensions
• delivery-modes.md — boss-primary / broadcast / generate-only
• formatting-standard.md — user-visible formatting rules
• failure-handling.md — status model and fallback behavior
• lifecycle-automation.md — archive-first lifecycle stages and recommended automation chain
• release-readiness.md — what is strong now vs what is still pre-release
• known-limitations.md — current gaps and intentionally disabled behavior
• validation-report.md — verified behaviors and remaining checks before public upload

assets/

• raw-template.md
• synthesized-template.md
• boss-summary-template.md
• index-template.json