Deadline Monitor
Use this when someone asks for a deadline / milestone risk digest or check-in for a GitHub repo
or GitHub organization (e.g. "看看 aevatarAI/aevatar 临近截止的 milestone,生成今日风险摘要",
"看看 ChronoAIProject 今天有哪些截止风险").
You (the agent) run the WHOLE flow yourself, calling your own tools — then actually post the
result. This skill hands you the step plan plus a deterministic payload builder so the Lark
message body is exact. Do NOT just describe what you would do, and do NOT refuse because "the
skill only formats" — that is wrong; drive the flow. Never ask the user for tokens; your NyxID
tools broker all credentials.
How to run it
Resolve inputs. Parse the target from the user's message and today's date. The target is
either:
- a concrete repo:
owner/name, for example aevatarAI/aevatar;
- a GitHub org/user only:
owner, for example ChronoAIProject.
Do not fabricate a repo from an org/user name. If the target has no slash, first treat it as an
org/user candidate and verify it through GitHub discovery. Never rewrite a single-segment target
into owner/name, and never call /repos/{owner}/{name}/... until discovery returns an actual
repository full_name.
Choose a mode: daily_risk_digest (open milestones with due dates, ranked by days left),
orphan_milestone_digest (open issues with no milestone), friday_progress_report
(milestone progress), or checkin_payloads (per-assignee check-ins). Default to
daily_risk_digest.
Collect from GitHub using your NyxID-brokered GitHub access (the api-github service via
nyxid_proxy).
For a concrete repo owner/name, call:
- milestones:
GET /repos/{owner}/{name}/milestones?state=open&sort=due_on&direction=asc
- issues:
GET /repos/{owner}/{name}/issues?state=open&per_page=100
For an org/user target owner, first discover repos:
- org repos:
GET /orgs/{owner}/repos?per_page=100&type=all
- if that returns 404, user repos:
GET /users/{owner}/repos?per_page=100&type=owner
Then call the repo endpoints above for each discovered non-archived repo you need. Keep the
number of repo fan-out calls reasonable: for daily_risk_digest and friday_progress_report,
milestone calls are enough; for orphan_milestone_digest and checkin_payloads, issue calls are
needed. Preserve the full repo name on every row (repo or repository, e.g.
ChronoAIProject/Ornn) before passing data to the builder. Keep only the fields you need
(title, due_on, state, closed_at, open/closed issue counts, html_url, assignee, milestone).
If every GitHub call for the selected target returns 404, stop and report that the target was not
found; do not build or send an empty "no risk" digest from failed calls.
Build the exact Lark payload deterministically. Run scripts/build_deadline_monitor_payload.js
via code_execute (language "javascript"). code_execute has NO stdin, so inline the collected
data into the code — the script exports a builder, so append
console.log(JSON.stringify(buildPayload({ mode: "<mode>", today: "<YYYY-MM-DD>", ...rows }))).
Use the lark.body (and message) from its output. If you genuinely cannot execute code,
construct the body by following references/deadline-monitor-contract.md field-for-field.
Send it to the current Lark chat. Post the digest into this conversation with
lark_messages_send (or lark_messages_reply to reply, or reply_with_interaction for a
card) using the lark.body from step 3.
Report a one-line summary of what you posted (mode, repo, item count).
If GitHub succeeds and returns no matching items for the chosen mode, say so plainly and stop — do
not invent records. If GitHub failed, report the failure instead of posting a clean empty digest.
Payload builder reference (step 3)
Modes the bundled script accepts:
daily_risk_digest — today + GitHub milestones (or precomputed risk rows).
orphan_milestone_digest — today + open GitHub issues.
friday_progress_report — today + milestone records.
checkin_payloads — selected check-in rows (after assignee mapping / filtering / dedupe).
Fields may be at the root or under body. Exact aliases, required keys, and examples are in
references/deadline-monitor-contract.md.
Script output is one of: skip: true; needs_more_information: true + missing keys; a digest
object with message, larkBody, and lark.body; or a check-in object with per-item
larkMsgBody, lark.body, githubCommentBody, and github.body. The Lark bodies already match
the field names Lark's send API expects.
Guardrails
- Only use data GitHub actually returned — never invent milestones, issues, user ids, or status.
- Prefer running the bundled script over hand-building JSON; it is the deterministic source of truth.
- All sending and token exchange go through your NyxID-brokered tools; do not handle raw secrets.
1---2name: deadline-monitor-digest-payload-builder3version: "2.3"4description: Runs the deadline-monitor flow end to end — reads GitHub repo or org milestones/issues, ranks deadline risk, builds the exact Lark digest with the bundled deterministic builder, and posts it to the current Lark chat.5metadata:6 source: https://skillmd.com/skills/chronoaiproject/deadline-monitor-digest-payload-builder7 category: plain8 tag:9 - aevatar10 - s-capability11 - deadline-monitor12 - lark13 - github14 - payload-builder15 clawdbot:16 emoji: "alarm_clock"17 files:18 - "references/*"19 - "scripts/*"20---2122# Deadline Monitor2324Use this when someone asks for a deadline / milestone risk digest or check-in for a GitHub repo25or GitHub organization (e.g. "看看 aevatarAI/aevatar 临近截止的 milestone,生成今日风险摘要",26"看看 ChronoAIProject 今天有哪些截止风险").2728**You (the agent) run the WHOLE flow yourself, calling your own tools — then actually post the29result.** This skill hands you the step plan plus a deterministic payload builder so the Lark30message body is exact. Do NOT just describe what you would do, and do NOT refuse because "the31skill only formats" — that is wrong; drive the flow. Never ask the user for tokens; your NyxID32tools broker all credentials.3334## How to run it35361. **Resolve inputs.** Parse the target from the user's message and today's date. The target is37 either:38 - a concrete repo: `owner/name`, for example `aevatarAI/aevatar`;39 - a GitHub org/user only: `owner`, for example `ChronoAIProject`.4041 Do not fabricate a repo from an org/user name. If the target has no slash, first treat it as an42 org/user candidate and verify it through GitHub discovery. Never rewrite a single-segment target43 into `owner/name`, and never call `/repos/{owner}/{name}/...` until discovery returns an actual44 repository `full_name`.4546 Choose a mode: `daily_risk_digest` (open milestones with due dates, ranked by days left),47 `orphan_milestone_digest` (open issues with no milestone), `friday_progress_report`48 (milestone progress), or `checkin_payloads` (per-assignee check-ins). Default to49 `daily_risk_digest`.50512. **Collect from GitHub** using your NyxID-brokered GitHub access (the `api-github` service via52 `nyxid_proxy`).5354 For a concrete repo `owner/name`, call:55 - milestones: `GET /repos/{owner}/{name}/milestones?state=open&sort=due_on&direction=asc`56 - issues: `GET /repos/{owner}/{name}/issues?state=open&per_page=100`5758 For an org/user target `owner`, first discover repos:59 - org repos: `GET /orgs/{owner}/repos?per_page=100&type=all`60 - if that returns 404, user repos: `GET /users/{owner}/repos?per_page=100&type=owner`6162 Then call the repo endpoints above for each discovered non-archived repo you need. Keep the63 number of repo fan-out calls reasonable: for `daily_risk_digest` and `friday_progress_report`,64 milestone calls are enough; for `orphan_milestone_digest` and `checkin_payloads`, issue calls are65 needed. Preserve the full repo name on every row (`repo` or `repository`, e.g.66 `ChronoAIProject/Ornn`) before passing data to the builder. Keep only the fields you need67 (title, due_on, state, closed_at, open/closed issue counts, html_url, assignee, milestone).6869 If every GitHub call for the selected target returns 404, stop and report that the target was not70 found; do not build or send an empty "no risk" digest from failed calls.71723. **Build the exact Lark payload deterministically.** Run `scripts/build_deadline_monitor_payload.js`73 via `code_execute` (language "javascript"). `code_execute` has NO stdin, so inline the collected74 data into the code — the script exports a builder, so append75 `console.log(JSON.stringify(buildPayload({ mode: "<mode>", today: "<YYYY-MM-DD>", ...rows })))`.76 Use the `lark.body` (and `message`) from its output. If you genuinely cannot execute code,77 construct the body by following `references/deadline-monitor-contract.md` field-for-field.78794. **Send it to the current Lark chat.** Post the digest into this conversation with80 `lark_messages_send` (or `lark_messages_reply` to reply, or `reply_with_interaction` for a81 card) using the `lark.body` from step 3.82835. **Report** a one-line summary of what you posted (mode, repo, item count).8485If GitHub succeeds and returns no matching items for the chosen mode, say so plainly and stop — do86not invent records. If GitHub failed, report the failure instead of posting a clean empty digest.8788## Payload builder reference (step 3)8990Modes the bundled script accepts:9192- `daily_risk_digest` — `today` + GitHub milestones (or precomputed risk rows).93- `orphan_milestone_digest` — `today` + open GitHub issues.94- `friday_progress_report` — `today` + milestone records.95- `checkin_payloads` — selected check-in rows (after assignee mapping / filtering / dedupe).9697Fields may be at the root or under `body`. Exact aliases, required keys, and examples are in98`references/deadline-monitor-contract.md`.99100Script output is one of: `skip: true`; `needs_more_information: true` + `missing` keys; a digest101object with `message`, `larkBody`, and `lark.body`; or a check-in object with per-item102`larkMsgBody`, `lark.body`, `githubCommentBody`, and `github.body`. The Lark bodies already match103the field names Lark's send API expects.104105## Guardrails106107- Only use data GitHub actually returned — never invent milestones, issues, user ids, or status.108- Prefer running the bundled script over hand-building JSON; it is the deterministic source of truth.109- All sending and token exchange go through your NyxID-brokered tools; do not handle raw secrets.