jbaruch/nanoclaw-conferences

Conference CFP discovery for NanoClaw. Finds open calls-for-papers relevant to the user across Java/AI/developer conferences, ranks them by relevance, and maintains persistent per-CFP state (sent / dismissed / remind) across sessions so the same conference is never surfaced twice.

Per-chat overlay tile. Install via NanoClaw's containerConfig.additionalTiles mechanism.

Capabilities

1. Multi-source discovery — Sessionize speaker API, developers.events, javaconferences.org, plus targeted web search for gaps
2. Source-aware verification — Sessionize is authority for Sessionize-sourced entries; non-Sessionize feeds are deadline-of-record; batched re-verification per run
3. AI relevance analysis — tiered routing (javaconferences.org auto-approve, blocklist, then AI judgement against the user's speaking topics and relevance criteria)
4. Persistent state — sent / dismissed / remind / approved / conflict status per CFP in cfp-state.json, immutable once the user acts on an entry
5. Travel-conflict detection — cross-checks CFP conference dates against the trip schedule and flags overlaps
6. Priority-interest tagging — deterministic prefilter plus AI arbitration against an owner-defined priority list

Installation

tessl install jbaruch/nanoclaw-conferences

Add to a chat's overlay tile list via update_group_config:

additionalTiles: ["nanoclaw-conferences"]

Load the overlay at the main or trusted tier — the skill reads /workspace/trusted/user_professional.md for the user's speaking topics.

Required environment

The CFP pipeline calls the Sessionize universal API deterministically from its own scripts (so the large responses never enter the agent's context — see jbaruch/nanoclaw-conferences#7/#9), which needs two host-injected keys in the container environment:

SESSIONIZE_API_BASE optionally overrides the https://sessionize.com/api/universal base (tests / proxying). These are the same keys the NanoClaw host already holds for its sessionize_* tools; the plugin reads them from the injected environment and bundles no secret of its own. Remaining data comes from the host fetch_markdown tool and public JSON feeds (developers.events, javaconferences.org); the sessionize_* MCP tools remain available for ad-hoc queries.

Runtime data

The skill reads and writes files under the shared /workspace/group/ mount:

Reads of admin-owned files resolve because admin co-loads with this overlay in the same chat via the shared mount; each is optional and degrades gracefully when absent.

Skills

Skill scripts

The skill bundle includes deterministic scripts the agent invokes from the SKILL.md steps:

• scripts/check-cfps-fetch.py — fetches + filters CFPs from the public feeds, applies the source-list and blocklist
• scripts/discover-open-cfps.py — Step 2 deterministic Sessionize open-CFP discovery via the speaker API (candidate shape and filter rules in its docstring)
• scripts/prepare-sessionize-batch.py — routes entries by effective source and derives Sessionize slugs for a single batched verification call
• scripts/apply-sessionize-results.py — joins batched Sessionize results back to entries and emits per-entry verdicts
• scripts/verify-sessionize.py — Step 5 verification driver: collapses prepare → live universal-API round-trip → apply into one invocation and writes the verify-evidence.json marker
• scripts/backfill-source.py — infers a missing source from the cfp_url host
• scripts/backfill-name.py — derives a fallback name for nameless records from the slug or cfp_url
• scripts/expire-cfps.py — expires stale non-Sessionize open/approved entries whose deadline passed
• scripts/match-priorities.py — deterministic priority-interest prefilter
• scripts/dedup-by-url.py — collapses entries whose cfp_url normalises to the same host+path, preserving the highest-priority source attribution and inheriting missing metadata from dropped copies
• scripts/commit-state.py — lock-owning committer for the Step 8 working-set write (the agent never writes cfp-state.json directly)
• scripts/stamp-schema-version.py — owner-side schema_version stamper for cfp-state.json
• scripts/stamp-last-checked.py — evidence-gated single writer of the _last_checked freshness heartbeat (exit 3 = verification not evidenced)
• scripts/run-state.py — per-run checkpoint store for resumable runs (begin/save/load/invalidate/done; schema in references/run-state.md)
• scripts/audit-sessionize-key-drift.py — reports Sessionize slug drift in stored state
• scripts/state_lock.py — shared advisory-flock module (not a CLI) that serialises every cfp-state.json mutator's read-modify-write

The nightly-cfp-sync cadence wrapper carries its own scripts:

• scripts/precheck-nightly-cfp-sync.py — fire-time precheck that gates wake-ups by the cadence cursor
• scripts/stamp-cursor.py — advances the nightly-cfp-sync-cursor.json success cursor after a clean run

Status

• V1.1 — adds the nightly-cfp-sync cadence wrapper alongside check-cfps, so the plugin owns both the user-driven CFP lookup and its scheduled refresh (mirrors nanoclaw-flight-assist bundling its sync-tripit cadence driver with check-travel-bookings). The wrapper materialises one scheduled_tasks row in chats that load this overlay.
• V1 — migrated check-cfps from nanoclaw-admin as a standalone per-chat overlay tile. Full multi-source discovery, source-aware Sessionize verification, AI relevance routing, persistent state with owner-side schema migration, travel-conflict detection, and priority-interest tagging.

See CHANGELOG.md for version history.