Write(clients/**) rules alone let a competitor scan's report write be denied.
Writes inside clients/, team/ and reports/ are now allowed by resolved path;
anything refused is logged with the tool and path.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Places API (New): competitor scans find local rivals on Google Maps and the
server reads real ratings/review counts by place id; client's own listing
fills the "You" row; recent reviews for praise/complaint themes; 30-day cache
- Google Trends via DataForSEO: rising/top related searches and % change per
keyword in the client's region; used by trend-scan and to check gap demand
- Instagram Hashtag Search: top/recent posts under a niche hashtag (tool +
Competitors page panel), counts Instagram's 30 tags/week limit
- Server settings → Research: keys plus on/off switches for each source
(hashtag search off until Meta approves Public Content Access)
- Reports mark website-sourced ratings as unverified; skill takes handles
only from a business's own site
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- competitor-scan skill + run type: web search for local rivals and
niche benchmarks, competitor_suggest checks each on Instagram
- Suggestions inbox on Competitors page (Track / Dismiss)
- Report blocks: leaderboard, size-vs-engagement map, format lift,
market's best posts, hashtag gap, Google ratings
- Stats: engagement by format, top hashtags; audit adds leaderboard
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Reports: mc: blocks (score, KPIs, reach, funnel, followers, formats,
rhythm, top posts, actions, screenshots, callouts) drawn as SVG from
stored account data, frozen per report; same page in PDF and preview
- Audit/weekly-report runs refresh account numbers first
- Screenshots tab: monthly shot list, uploads, examples; audit reads them
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Matches the laptop Ollama setup (num_ctx 65536 on the model): requests
stay readable within the agent's ~5-minute wait while big tool results
still fit.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Local models (Ollama on a laptop GPU) break on requests over ~19k tokens.
Model-server runs now send only the tools and skill the job needs and
summarise the conversation at ~16k tokens. Publish, schedule, reply and
DM tools are removed from every run.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With a model server address set, runs send every model slot (opus/sonnet/haiku/subagent)
to the chosen model and never send the Anthropic key there. Tested locally with
qwen3.6:35b-a3b through the Agent SDK.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Dashboard runs couldn't call plan_add, so weekly plans never reached the Calendar.
Adds the calendar tools (plan items only, nothing is scheduled) and the read tools
audits use (competitors, demographics, mentions, connections, library, brand kit).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Dashboard (content library, publishing queue with approvals, inbox, team
management, AI runs), Meta connector, branded PDF reports, agent skills,
and Docker packaging (Dockerfile, docker-compose with Caddy HTTPS,
backup and path-relocation scripts).
Client data, live data and secrets are excluded via .gitignore.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
