gooseworks-ai/goose-skills / skills/marketing/research-competitors
research-competitors
VerifiedCompetitor research: what every main rival is running, launching, charging and saying now, what works for them, and what that means for you. Use when the marketer asks for competitor research or a teardown, "what ads are my competitors running", "how do they price", "find gaps and angles", "what changed this week" (a weekly watch), or "X just launched / shut down / cut prices" (react now). Also builds a sales battlecard, checks a competitor page's tools/pixels, or studies who engages with their LinkedIn posts. Delivers a readable report saved to the brand, so the next run reports what changed. Never copies their words or disparages them. Not for what people say about you or your market — use research-market; not for who your buyers are — use research-audience.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 11.7 KB |
| references/angles-and-gaps.md | markdown | 7.4 KB |
| references/battlecard.md | markdown | 3.8 KB |
| references/focused-outputs.md | markdown | 2.7 KB |
| references/linkedin-engagers.md | markdown | 6.3 KB |
| references/page-technology.md | markdown | 2.7 KB |
| references/reading-ads-honestly.md | markdown | 8.6 KB |
| references/report-format.md | markdown | 10.0 KB |
| references/research-spec.md | markdown | 34.5 KB |
| references/rival-craft.md | markdown | 16.0 KB |
| references/run-memory.md | markdown | 6.6 KB |
| references/watch-and-react.md | markdown | 12.7 KB |
| skill.meta.json | config | 5.5 KB |
| tests/examples/2026-10-22/battlecard.md | markdown | 5.5 KB |
| tests/examples/2026-10-22/engagement-report.md | markdown | 3.7 KB |
| tests/examples/2026-10-22/engagers.csv | other | 1.7 KB |
| tests/examples/2026-10-22/excluded.csv | other | 1.0 KB |
| tests/examples/2026-10-22/technology.md | markdown | 3.2 KB |
| tests/fixtures/engagers.csv | other | 1.9 KB |
| tests/fixtures/focused-evidence.md | markdown | 3.8 KB |
| tests/fixtures/page-source.html | other | 614 B |
| tests/test-by-hand.md | markdown | 7.5 KB |
references/run-memory.md
View on GitHub# Run memory for research-competitors
What this playbook hands the `run-memory` skill (`../../run-memory/SKILL.md`) at the start and the end of every run. The atom does the work (find the last record, check recommendations, write "since last time", save the new record); this file says what is specific here.
## Jobs
| `job` | When | `cadence` |
|---|---|---|
| `competitor research` | a full run — on ask, or the quarterly teardown | `quarterly` (on ask: `ad hoc`) |
| `competitor watch` | the what-changed run — weekly scheduled, or "what changed this week?" | `weekly` |
Focused jobs: `battlecard`, `page technology`, `LinkedIn engagement` use `cadence: ad hoc` and `subject: <rival-slug>`. Each reads only its own job/subject record, not the full teardown baseline. Record output links, source dates, proof/approval gaps and 3–6 useful counts; engagement records contain counts and post URLs only, no people or comment text. They do not update the full teardown snapshots unless those fields were actually checked.
For the full research/watch: `playbook` = `research-competitors`; no `subject` (one record covers all rivals). **The last record is the newest record of either job** — a watch after a quarterly teardown reads the teardown (its fresh ad counts and new recommendations), and a teardown reads the last watch. Each run saves under its own job.
**No run record, but an earlier report exists** (saved before run memory): say "first tracked run; last report <date>", take that report's baselines table as the before (labelled "from the <date> report"), and log its "What to do" moves as recommendations with new ids, each status taken from a source. **Nothing at all:** "first run" — no "since last time"; this run's numbers are the baseline.
**What it does not replace:** each rival's record (`research/competitors/<slug>.md`) and its **Snapshot** block stay the source of truth for per-rival state — ad ids and start dates, pricing text, headline, open roles. The run record keeps only the numbers below and links the rival records by path; it never copies a snapshot.
## Numbers (the record's "Numbers" section)
One row per main rival, plus a row for us. Each cell carries its read date and how it was read; every number is `public page` (as read, not verified), except our own figures from the brand's data.
| Number | How it's read | Unit / note |
|---|---|---|
| Active ads per library (Meta, LinkedIn, Google, TikTok) | the library page, by advertiser | a count labelled with what it counts (Meta "active now", Google "shown since <date>") — never 0 unless the library loaded; compare a library only with itself across runs |
| Longest-running active ad | the library | days, with its ad id |
| Entry price and a typical buyer's monthly cost | pricing page (scenario from rival-craft §6) | $/month; "not public" |
| Launches in the last 90 days | changelog, news | count, with the biggest named |
| Reviews on the main review site | the review page | count and rating, with the site |
| Open roles | careers page | count, biggest team |
| Us: rank for "<main rival> alternative" | web search, signed out | position in the top 20, or "not in top 20" |
| Us: AI answers that name us | the 3–4 category questions | n of N |
The watch's thresholds (`watch-and-react.md`) are judged against these numbers and the rival snapshots, never against memory.
## Recommendations and how to check them
The report's "What to do" moves become the record's recommendations: id, owner, review-by date and "how we'll know". **Owner mapping:** the founder, "whoever runs the website", the ad or content maker → `marketer` (or the maker playbook once it takes the move); waiting for a yes in GooseWorks → `approval page`; this job's next run → `agent`.
**Not yet due:** a recommendation whose review-by date hasn't come, with nothing showing yet, stays **open** — carried forward unchanged and never asked about, with status `open` and its review-by date. Typical tests, with their **baseline recipe** (what to re-read, the same way, next time):
- **A page or post we publish** (comparison, alternatives, pricing explainer, a blog post): store the planned page title and slug in "how we'll know". Check: the brand's `/sitemap.xml` and blog feed for pages added since the recommendation's date, then fetch the match; rank for "<rival> alternative" in the top 20 by <date + 8 weeks> — the same signed-out search. Ranking moves slowly: never judge a page under 6 weeks old.
- **An ad or post on an angle** (owner: the ad or content maker): exists by <date> — `gooseworks:social_post_list`, `gooseworks:campaign_read`, or `gooseworks:meta_read` for ads GooseWorks launched; its result belongs to analyze-performance — link its report, don't re-judge here.
- **A rival tracked or dropped:** `gooseworks:competitor_read`.
- **A move we decided not to make** ("What we're not doing") is recorded as a decision, not a recommendation.
`outcome_sources` = the brand's own sitemap, blog feed and pages (fetch), `social_post_list`, `campaign_read`, `meta_read` (GooseWorks-launched ads only), `competitor_read`, analyze-performance's latest report.
## The founder's answers
Save the run record **before** the closing message, and put the one-line "saved" note into that message. When the founder answers a decision in the same conversation, re-save the same day's record under the same title (keep-history replaces only that exact title) with the Decisions and owners updated. A later run treats a question as answered only from a dated Decision in a record.
## Scheduled watch (nobody present)
Run-memory's scheduled rules apply in full. Here specifically:
- Free reads only (news, pages and feeds a fetch can read, the server's free samples); never the ad libraries — the record says "ad libraries not read (scheduled run)" under Not checked, and the ad-count cells keep the previous value marked "carried, not re-read".
- **Nothing crossed a threshold** → no report is sent, but the run record is still written: the numbers re-read, "nothing crossed a threshold", the open recommendations carried forward. That record is the next baseline.
- **Something crossed** → a what-changed report through write-report: what was re-read, what changed against the stored numbers, then "So, do this first (because of …)", in "In short"; its "Decisions for you" holds any `unknown` statuses (once each, with the date asked).
- **Questions on a quiet week** (no report goes out): mark the item `to ask` in the record with "asked on" empty; it is asked in the next report or the next run with the founder present.
- A run that can't read its sources (a connection missing, every page failing) writes a `Not run: <why>` record and keeps the previous baseline.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| add-product-e2e-evalpaperclipai/paperclipAdd or extend a Paperclip full-stack runner E2E workflow, fixture, matcher, or report evidence path for local or Daytona execution. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| add-runner-evalpaperclipai/paperclipAdd or extend a Paperclip Runner protocol evaluation definition, roster, assertion, or report fixture with provenance and narrow validation. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| diagnose-why-work-stoppedpaperclipai/paperclipCompatibility entry point for failed, stalled, looping, or over-recovered Paperclip work. Use fix-failed-paperclip-run for the full diagnosis, regression, repair, real verification, and green-PR workflow; honor an explicitly requested diagnosis-only scope. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| fix-failed-paperclip-runpaperclipai/paperclipDiagnose and fix failed, stuck, slow, or incorrectly blocked Paperclip tasks and runs. Use with a task/run URL or failure report to trace the root cause, prove a regression red then green, check second-order effects, verify the real workflow, and deliver a PR with passing checks. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |